d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
01646adbc8
freebsd-dev/sys/dev/isp/isp_pci.c

2507 lines
64 KiB
C
Raw Normal View History

Use __FBSDID(). Also some minor style cleanups.
2003-08-24 17:55:58 +00:00
/*-
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
*
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
* Copyright (c) 1997-2006 by Matthew Jacob
* All rights reserved.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice immediately at the beginning of the file, without modification,
* this list of conditions, and the following disclaimer.
some copyright cleanups
2000-09-21 20:16:04 +00:00
* 2. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
*
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
/*
* PCI specific probe and attach routines for Qlogic ISP SCSI adapters.
* FreeBSD Version.
*/
Use __FBSDID(). Also some minor style cleanups.
2003-08-24 17:55:58 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/module.h>
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
#if __FreeBSD_version >= 700000
#include <sys/linker.h>
#include <sys/firmware.h>
#endif
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
#include <sys/bus.h>
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#if __FreeBSD_version < 500000
#include <pci/pcireg.h>
#include <pci/pcivar.h>
#include <machine/bus_memio.h>
#include <machine/bus_pio.h>
#else
Use uintmax_t and %j to print bus dma segment members rather than casting to long long and using %ll.
2005-12-15 22:12:27 +00:00
#include <sys/stdint.h>
Prefer new location of pci include files (which have only been in the tree for two or more years now), except in a few places where there's code to be compatible with older versions of FreeBSD.
2003-08-22 06:00:27 +00:00
#include <dev/pci/pcireg.h>
#include <dev/pci/pcivar.h>
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#endif
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
#include <machine/bus.h>
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
#include <machine/resource.h>
#include <sys/rman.h>
#include <sys/malloc.h>
#include <dev/isp/isp_freebsd.h>
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#if __FreeBSD_version < 500000
#define BUS_PROBE_DEFAULT 0
#endif
static uint16_t isp_pci_rd_reg(ispsoftc_t *, int);
static void isp_pci_wr_reg(ispsoftc_t *, int, uint16_t);
static uint16_t isp_pci_rd_reg_1080(ispsoftc_t *, int);
static void isp_pci_wr_reg_1080(ispsoftc_t *, int, uint16_t);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_rd_isr(ispsoftc_t *, uint16_t *, uint16_t *, uint16_t *);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_rd_isr_2300(ispsoftc_t *, uint16_t *, uint16_t *, uint16_t *);
static int isp_pci_mbxdma(ispsoftc_t *);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_dmasetup(ispsoftc_t *, XS_T *, ispreq_t *, uint16_t *, uint16_t);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_dmateardown(ispsoftc_t *, XS_T *, uint16_t);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
static void isp_pci_reset1(ispsoftc_t *);
static void isp_pci_dumpregs(ispsoftc_t *, const char *);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
static struct ispmdvec mdvec = {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
isp_pci_rd_isr,
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
isp_pci_rd_reg,
isp_pci_wr_reg,
isp_pci_mbxdma,
isp_pci_dmasetup,
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
isp_pci_dmateardown,
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
NULL,
isp_pci_reset1,
isp_pci_dumpregs,
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
NULL,
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
BIU_BURST_ENABLE|BIU_PCI_CONF1_FIFO_64
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
};
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
static struct ispmdvec mdvec_1080 = {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
isp_pci_rd_isr,
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
isp_pci_rd_reg_1080,
isp_pci_wr_reg_1080,
isp_pci_mbxdma,
isp_pci_dmasetup,
isp_pci_dmateardown,
NULL,
isp_pci_reset1,
isp_pci_dumpregs,
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
NULL,
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
BIU_BURST_ENABLE|BIU_PCI_CONF1_FIFO_64
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
};
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
static struct ispmdvec mdvec_12160 = {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
isp_pci_rd_isr,
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
isp_pci_rd_reg_1080,
isp_pci_wr_reg_1080,
isp_pci_mbxdma,
isp_pci_dmasetup,
isp_pci_dmateardown,
NULL,
isp_pci_reset1,
isp_pci_dumpregs,
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
NULL,
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
BIU_BURST_ENABLE|BIU_PCI_CONF1_FIFO_64
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
};
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
static struct ispmdvec mdvec_2100 = {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
isp_pci_rd_isr,
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
isp_pci_rd_reg,
isp_pci_wr_reg,
isp_pci_mbxdma,
isp_pci_dmasetup,
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
isp_pci_dmateardown,
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
NULL,
isp_pci_reset1,
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
isp_pci_dumpregs
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
};
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
static struct ispmdvec mdvec_2200 = {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
isp_pci_rd_isr,
isp_pci_rd_reg,
isp_pci_wr_reg,
isp_pci_mbxdma,
isp_pci_dmasetup,
isp_pci_dmateardown,
NULL,
isp_pci_reset1,
isp_pci_dumpregs
};
static struct ispmdvec mdvec_2300 = {
isp_pci_rd_isr_2300,
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
isp_pci_rd_reg,
isp_pci_wr_reg,
isp_pci_mbxdma,
isp_pci_dmasetup,
isp_pci_dmateardown,
NULL,
isp_pci_reset1,
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
isp_pci_dumpregs
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
};
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
#ifndef PCIM_CMD_INVEN
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
#define PCIM_CMD_INVEN 0x10
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
#endif
#ifndef PCIM_CMD_BUSMASTEREN
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
#define PCIM_CMD_BUSMASTEREN 0x0004
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
#endif
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
#ifndef PCIM_CMD_PERRESPEN
#define PCIM_CMD_PERRESPEN 0x0040
#endif
#ifndef PCIM_CMD_SEREN
#define PCIM_CMD_SEREN 0x0100
#endif
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
#ifndef PCIM_CMD_INTX_DISABLE
#define PCIM_CMD_INTX_DISABLE 0x0400
#endif
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
#ifndef PCIR_COMMAND
#define PCIR_COMMAND 0x04
#endif
#ifndef PCIR_CACHELNSZ
#define PCIR_CACHELNSZ 0x0c
#endif
#ifndef PCIR_LATTIMER
#define PCIR_LATTIMER 0x0d
#endif
Read the board revision and trim cache line size back from 16 to 1 for early revision 2100 boards. Make sure to turn ROM off for these boards.
1999-04-04 01:14:02 +00:00
#ifndef PCIR_ROMADDR
#define PCIR_ROMADDR 0x30
#endif
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
#ifndef PCI_VENDOR_QLOGIC
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
#define PCI_VENDOR_QLOGIC 0x1077
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
#endif
#ifndef PCI_PRODUCT_QLOGIC_ISP1020
#define PCI_PRODUCT_QLOGIC_ISP1020 0x1020
#endif
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP1080
#define PCI_PRODUCT_QLOGIC_ISP1080 0x1080
#endif
This should enable 10160 support. As best as I can tell, the same f/w as 12160 is used, and otherwise, this is just a single channel variant of the 10160. MFC after: 0 days
2002-10-11 17:28:01 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP10160
#define PCI_PRODUCT_QLOGIC_ISP10160 0x1016
#endif
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP12160
#define PCI_PRODUCT_QLOGIC_ISP12160 0x1216
#endif
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP1240
#define PCI_PRODUCT_QLOGIC_ISP1240 0x1240
#endif
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP1280
#define PCI_PRODUCT_QLOGIC_ISP1280 0x1280
#endif
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP2100
#define PCI_PRODUCT_QLOGIC_ISP2100 0x2100
#endif
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP2200
#define PCI_PRODUCT_QLOGIC_ISP2200 0x2200
#endif
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP2300
#define PCI_PRODUCT_QLOGIC_ISP2300 0x2300
#endif
#ifndef PCI_PRODUCT_QLOGIC_ISP2312
#define PCI_PRODUCT_QLOGIC_ISP2312 0x2312
#endif
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP2322
#define PCI_PRODUCT_QLOGIC_ISP2322 0x2322
#endif
Put in at least an attempt to ID the 2422 (4Gb part)
2006-01-26 05:04:35 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP2422
#define PCI_PRODUCT_QLOGIC_ISP2422 0x2422
#endif
Support the DELL OEM 2312 cards (1077,6312). Many thanks to Stormweb for making the h/w available for testing. MFC after: 2 days
2005-01-23 06:23:55 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP6312
#define PCI_PRODUCT_QLOGIC_ISP6312 0x6312
#endif
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
#ifndef PCI_PRODUCT_QLOGIC_ISP6322
#define PCI_PRODUCT_QLOGIC_ISP6322 0x6322
#endif
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
#define PCI_QLOGIC_ISP1020 \
((PCI_PRODUCT_QLOGIC_ISP1020 << 16) | PCI_VENDOR_QLOGIC)
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
#define PCI_QLOGIC_ISP1080 \
((PCI_PRODUCT_QLOGIC_ISP1080 << 16) | PCI_VENDOR_QLOGIC)
This should enable 10160 support. As best as I can tell, the same f/w as 12160 is used, and otherwise, this is just a single channel variant of the 10160. MFC after: 0 days
2002-10-11 17:28:01 +00:00
#define PCI_QLOGIC_ISP10160 \
((PCI_PRODUCT_QLOGIC_ISP10160 << 16) | PCI_VENDOR_QLOGIC)
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
#define PCI_QLOGIC_ISP12160 \
((PCI_PRODUCT_QLOGIC_ISP12160 << 16) | PCI_VENDOR_QLOGIC)
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
#define PCI_QLOGIC_ISP1240 \
((PCI_PRODUCT_QLOGIC_ISP1240 << 16) | PCI_VENDOR_QLOGIC)
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
#define PCI_QLOGIC_ISP1280 \
((PCI_PRODUCT_QLOGIC_ISP1280 << 16) | PCI_VENDOR_QLOGIC)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
#define PCI_QLOGIC_ISP2100 \
((PCI_PRODUCT_QLOGIC_ISP2100 << 16) | PCI_VENDOR_QLOGIC)
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
#define PCI_QLOGIC_ISP2200 \
((PCI_PRODUCT_QLOGIC_ISP2200 << 16) | PCI_VENDOR_QLOGIC)
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
#define PCI_QLOGIC_ISP2300 \
((PCI_PRODUCT_QLOGIC_ISP2300 << 16) | PCI_VENDOR_QLOGIC)
#define PCI_QLOGIC_ISP2312 \
((PCI_PRODUCT_QLOGIC_ISP2312 << 16) | PCI_VENDOR_QLOGIC)
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
#define PCI_QLOGIC_ISP2322 \
((PCI_PRODUCT_QLOGIC_ISP2322 << 16) | PCI_VENDOR_QLOGIC)
oops
2006-01-26 06:15:58 +00:00
#define PCI_QLOGIC_ISP2422 \
((PCI_PRODUCT_QLOGIC_ISP2422 << 16) | PCI_VENDOR_QLOGIC)
Support the DELL OEM 2312 cards (1077,6312). Many thanks to Stormweb for making the h/w available for testing. MFC after: 2 days
2005-01-23 06:23:55 +00:00
#define PCI_QLOGIC_ISP6312 \
((PCI_PRODUCT_QLOGIC_ISP6312 << 16) | PCI_VENDOR_QLOGIC)
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
#define PCI_QLOGIC_ISP6322 \
((PCI_PRODUCT_QLOGIC_ISP6322 << 16) | PCI_VENDOR_QLOGIC)
Per msmith's request, don't attach to Qlogic 12160 id'd cards that have a certain SubVendorID.
2000-09-07 20:27:40 +00:00
/*
* Odd case for some AMI raid cards... We need to *not* attach to this.
*/
#define AMI_RAID_SUBVENDOR_ID 0x101e
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
#define IO_MAP_REG 0x10
#define MEM_MAP_REG 0x14
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
#define PCI_DFLT_LTNCY 0x40
#define PCI_DFLT_LNSZ 0x10
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
static int isp_pci_probe (device_t);
static int isp_pci_attach (device_t);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
struct isp_pcisoftc {
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t pci_isp;
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_t pci_dev;
struct resource * pci_reg;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
bus_space_tag_t pci_st;
bus_space_handle_t pci_sh;
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
void * ih;
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
int16_t pci_poff[_NREG_BLKS];
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_dma_tag_t dmat;
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
bus_dmamap_t *dmaps;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
};
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
static device_method_t isp_pci_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, isp_pci_probe),
DEVMETHOD(device_attach, isp_pci_attach),
{ 0, 0 }
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
};
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
static void isp_pci_intr(void *);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
static driver_t isp_pci_driver = {
"isp", isp_pci_methods, sizeof (struct isp_pcisoftc)
};
static devclass_t isp_devclass;
DRIVER_MODULE(isp, pci, isp_pci_driver, isp_devclass, 0, 0);
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
#if __FreeBSD_version >= 700000
MODULE_DEPEND(isp, ispfw, 1, 1, 1);
MODULE_DEPEND(isp, firmware, 1, 1, 1);
#else
extern ispfwfunc *isp_get_firmware_p;
#endif
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
static int
isp_pci_probe(device_t dev)
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
{
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
switch ((pci_get_device(dev) << 16) | (pci_get_vendor(dev))) {
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
case PCI_QLOGIC_ISP1020:
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_set_desc(dev, "Qlogic ISP 1020/1040 PCI SCSI Adapter");
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
break;
case PCI_QLOGIC_ISP1080:
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_set_desc(dev, "Qlogic ISP 1080 PCI SCSI Adapter");
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
break;
case PCI_QLOGIC_ISP1240:
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_set_desc(dev, "Qlogic ISP 1240 PCI SCSI Adapter");
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
break;
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
case PCI_QLOGIC_ISP1280:
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_set_desc(dev, "Qlogic ISP 1280 PCI SCSI Adapter");
break;
This should enable 10160 support. As best as I can tell, the same f/w as 12160 is used, and otherwise, this is just a single channel variant of the 10160. MFC after: 0 days
2002-10-11 17:28:01 +00:00
case PCI_QLOGIC_ISP10160:
device_set_desc(dev, "Qlogic ISP 10160 PCI SCSI Adapter");
break;
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
case PCI_QLOGIC_ISP12160:
Per msmith's request, don't attach to Qlogic 12160 id'd cards that have a certain SubVendorID.
2000-09-07 20:27:40 +00:00
if (pci_get_subvendor(dev) == AMI_RAID_SUBVENDOR_ID) {
return (ENXIO);
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_set_desc(dev, "Qlogic ISP 12160 PCI SCSI Adapter");
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
break;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
case PCI_QLOGIC_ISP2100:
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_set_desc(dev, "Qlogic ISP 2100 PCI FC-AL Adapter");
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
break;
add ISP_DISABLE_2200_SUPPORT defines; Add reference to 2200 F/W
1999-07-05 20:24:46 +00:00
case PCI_QLOGIC_ISP2200:
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_set_desc(dev, "Qlogic ISP 2200 PCI FC-AL Adapter");
add ISP_DISABLE_2200_SUPPORT defines; Add reference to 2200 F/W
1999-07-05 20:24:46 +00:00
break;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
case PCI_QLOGIC_ISP2300:
device_set_desc(dev, "Qlogic ISP 2300 PCI FC-AL Adapter");
break;
case PCI_QLOGIC_ISP2312:
device_set_desc(dev, "Qlogic ISP 2312 PCI FC-AL Adapter");
break;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
case PCI_QLOGIC_ISP2322:
device_set_desc(dev, "Qlogic ISP 2322 PCI FC-AL Adapter");
break;
Don't attach 2422's yet. It just confuses everyone.
2006-07-14 05:16:00 +00:00
#if 0
Put in at least an attempt to ID the 2422 (4Gb part)
2006-01-26 05:04:35 +00:00
case PCI_QLOGIC_ISP2422:
device_set_desc(dev, "Qlogic ISP 2422 PCI FC-AL Adapter");
break;
Don't attach 2422's yet. It just confuses everyone.
2006-07-14 05:16:00 +00:00
#endif
Support the DELL OEM 2312 cards (1077,6312). Many thanks to Stormweb for making the h/w available for testing. MFC after: 2 days
2005-01-23 06:23:55 +00:00
case PCI_QLOGIC_ISP6312:
device_set_desc(dev, "Qlogic ISP 6312 PCI FC-AL Adapter");
break;
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
case PCI_QLOGIC_ISP6322:
device_set_desc(dev, "Qlogic ISP 6322 PCI FC-AL Adapter");
break;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
default:
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
return (ENXIO);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
'Support' for ISP SBus cards. This code does not imply that SBus cards work yet. They hang for me. But I can't netboot the latest snapshot on my ultra1e, and things hang at bus_setup_intr time. Since I'm offline for a while, I thought I'd toss this in in case somebody else who has a bit better luck wants to fart around with it. Please try and wait until I get back to check things in.
2002-07-11 03:25:04 +00:00
if (isp_announced == 0 && bootverbose) {
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
printf("Qlogic ISP Driver, FreeBSD Version %d.%d, "
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
"Core Version %d.%d\n",
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
ISP_PLATFORM_VERSION_MAJOR, ISP_PLATFORM_VERSION_MINOR,
ISP_CORE_VERSION_MAJOR, ISP_CORE_VERSION_MINOR);
'Support' for ISP SBus cards. This code does not imply that SBus cards work yet. They hang for me. But I can't netboot the latest snapshot on my ultra1e, and things hang at bus_setup_intr time. Since I'm offline for a while, I thought I'd toss this in in case somebody else who has a bit better luck wants to fart around with it. Please try and wait until I get back to check things in.
2002-07-11 03:25:04 +00:00
isp_announced++;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
/*
* XXXX: Here is where we might load the f/w module
* XXXX: (or increase a reference count to it).
*/
Use BUS_PROBE_DEFAULT for pci probe return value
2005-03-05 18:17:35 +00:00
return (BUS_PROBE_DEFAULT);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#if __FreeBSD_version < 500000
static void
isp_get_options(device_t dev, ispsoftc_t *isp)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
{
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
uint64_t wwn;
int bitmap, unit;
unit = device_get_unit(dev);
if (getenv_int("isp_disable", &bitmap)) {
if (bitmap & (1 << unit)) {
isp->isp_osinfo.disabled = 1;
return;
}
}
if (getenv_int("isp_no_fwload", &bitmap)) {
if (bitmap & (1 << unit))
isp->isp_confopts |= ISP_CFG_NORELOAD;
}
if (getenv_int("isp_fwload", &bitmap)) {
if (bitmap & (1 << unit))
isp->isp_confopts &= ~ISP_CFG_NORELOAD;
}
if (getenv_int("isp_no_nvram", &bitmap)) {
if (bitmap & (1 << unit))
isp->isp_confopts |= ISP_CFG_NONVRAM;
}
if (getenv_int("isp_nvram", &bitmap)) {
if (bitmap & (1 << unit))
isp->isp_confopts &= ~ISP_CFG_NONVRAM;
}
if (getenv_int("isp_fcduplex", &bitmap)) {
if (bitmap & (1 << unit))
isp->isp_confopts |= ISP_CFG_FULL_DUPLEX;
}
if (getenv_int("isp_no_fcduplex", &bitmap)) {
if (bitmap & (1 << unit))
isp->isp_confopts &= ~ISP_CFG_FULL_DUPLEX;
}
if (getenv_int("isp_nport", &bitmap)) {
if (bitmap & (1 << unit))
isp->isp_confopts |= ISP_CFG_NPORT;
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
/*
* Because the resource_*_value functions can neither return
* 64 bit integer values, nor can they be directly coerced
* to interpret the right hand side of the assignment as
* you want them to interpret it, we have to force WWN
* hint replacement to specify WWN strings with a leading
* 'w' (e..g w50000000aaaa0001). Sigh.
*/
if (getenv_quad("isp_portwwn", &wwn)) {
isp->isp_osinfo.default_port_wwn = wwn;
isp->isp_confopts |= ISP_CFG_OWNWWPN;
}
if (isp->isp_osinfo.default_port_wwn == 0) {
isp->isp_osinfo.default_port_wwn = 0x400000007F000009ull;
}
if (getenv_quad("isp_nodewwn", &wwn)) {
isp->isp_osinfo.default_node_wwn = wwn;
isp->isp_confopts |= ISP_CFG_OWNWWNN;
}
if (isp->isp_osinfo.default_node_wwn == 0) {
isp->isp_osinfo.default_node_wwn = 0x400000007F000009ull;
}
bitmap = 0;
(void) getenv_int("isp_debug", &bitmap);
if (bitmap) {
isp->isp_dblev = bitmap;
} else {
isp->isp_dblev = ISP_LOGWARN|ISP_LOGERR;
}
if (bootverbose) {
isp->isp_dblev |= ISP_LOGCONFIG|ISP_LOGINFO;
}
#ifdef ISP_FW_CRASH_DUMP
bitmap = 0;
if (getenv_int("isp_fw_dump_enable", &bitmap)) {
if (bitmap & (1 << unit) {
size_t amt = 0;
if (IS_2200(isp)) {
amt = QLA2200_RISC_IMAGE_DUMP_SIZE;
} else if (IS_23XX(isp)) {
amt = QLA2300_RISC_IMAGE_DUMP_SIZE;
}
if (amt) {
FCPARAM(isp)->isp_dump_data =
malloc(amt, M_DEVBUF, M_WAITOK);
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
memset(FCPARAM(isp)->isp_dump_data, 0, amt);
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
} else {
device_printf(dev,
"f/w crash dumps not supported for card\n");
}
}
}
#endif
Fix RELENG_4 code version- isp_roles wasn't getting initialized so it ended up defaulting to ISP_ROLE_NONE. My testing hadn't caught it because I was deliberatly setting role via ioctl. Thanks to user Toni for lending me an alpha to test this on. MFC after: 0 days
2006-08-21 00:46:10 +00:00
bitmap = 0;
if (getenv_int("role", &bitmap)) {
isp->isp_role = bitmap;
} else {
isp->isp_role = ISP_DEFAULT_ROLES;
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
}
static void
isp_get_pci_options(device_t dev, int *m1, int *m2)
{
int bitmap;
int unit = device_get_unit(dev);
*m1 = PCIM_CMD_MEMEN;
*m2 = PCIM_CMD_PORTEN;
if (getenv_int("isp_mem_map", &bitmap)) {
if (bitmap & (1 << unit)) {
*m1 = PCIM_CMD_MEMEN;
*m2 = PCIM_CMD_PORTEN;
}
}
bitmap = 0;
if (getenv_int("isp_io_map", &bitmap)) {
if (bitmap & (1 << unit)) {
*m1 = PCIM_CMD_PORTEN;
*m2 = PCIM_CMD_MEMEN;
}
}
}
#else
static void
isp_get_options(device_t dev, ispsoftc_t *isp)
{
int tval;
const char *sptr;
add in a boot environment isp_disable flag
1999-07-06 01:24:20 +00:00
/*
* Figure out if we're supposed to skip this one.
*/
Eliminate the use of the getenv_int stuff we'd been using (with a bitmap for selecting unit). Instead, use the resource hints mechanism. One unfortunate situation here is that there is no resource_quad_value function- which is what I needed for WWN boot time replacement. Worse- you can't store the hint as just plain hint.isp.0.nodewwn="0x50000000aaaa0001" because this gets interpreted as an int- incorrectly because it can't be converted to an int. I can't even get this as a string. To work around this particular case for nodewwn && portwwn setting, this rather grotesque form will be used: hint.isp.0.nodewwn="w50000000aaaa0001" hint.isp.0.portwwn="w50000000aaaa0002" At the same time, if we have no hinted WWN, set the default WWN (which, btw, gets overridden if the card has valid NVRAM, which is usual) to 0x400000007F000009ull (which translates to NAA == IPv4, 127.0.0.9). Eliminate more printf's and replace them either with device_printf or isp_prt calls.
2001-03-01 02:21:36 +00:00
tval = 0;
if (resource_int_value(device_get_name(dev), device_get_unit(dev),
"disable", &tval) == 0 && tval) {
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
device_printf(dev, "disabled at user request\n");
isp->isp_osinfo.disabled = 1;
return;
Shuffle around how we do isp_disable management- make sure we return 0 so the unit number doesn't get reused. Make sure that if we've compiled for ISP_TARGET_MODE we set the default role to be ISP_ROLE_INITIATOR|ISP_ROLE_TARGET. Do some misc other cleanups.
2001-02-11 03:53:23 +00:00
}
Eliminate the use of the getenv_int stuff we'd been using (with a bitmap for selecting unit). Instead, use the resource hints mechanism. One unfortunate situation here is that there is no resource_quad_value function- which is what I needed for WWN boot time replacement. Worse- you can't store the hint as just plain hint.isp.0.nodewwn="0x50000000aaaa0001" because this gets interpreted as an int- incorrectly because it can't be converted to an int. I can't even get this as a string. To work around this particular case for nodewwn && portwwn setting, this rather grotesque form will be used: hint.isp.0.nodewwn="w50000000aaaa0001" hint.isp.0.portwwn="w50000000aaaa0002" At the same time, if we have no hinted WWN, set the default WWN (which, btw, gets overridden if the card has valid NVRAM, which is usual) to 0x400000007F000009ull (which translates to NAA == IPv4, 127.0.0.9). Eliminate more printf's and replace them either with device_printf or isp_prt calls.
2001-03-01 02:21:36 +00:00
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
tval = -1;
Eliminate the use of the getenv_int stuff we'd been using (with a bitmap for selecting unit). Instead, use the resource hints mechanism. One unfortunate situation here is that there is no resource_quad_value function- which is what I needed for WWN boot time replacement. Worse- you can't store the hint as just plain hint.isp.0.nodewwn="0x50000000aaaa0001" because this gets interpreted as an int- incorrectly because it can't be converted to an int. I can't even get this as a string. To work around this particular case for nodewwn && portwwn setting, this rather grotesque form will be used: hint.isp.0.nodewwn="w50000000aaaa0001" hint.isp.0.portwwn="w50000000aaaa0002" At the same time, if we have no hinted WWN, set the default WWN (which, btw, gets overridden if the card has valid NVRAM, which is usual) to 0x400000007F000009ull (which translates to NAA == IPv4, 127.0.0.9). Eliminate more printf's and replace them either with device_printf or isp_prt calls.
2001-03-01 02:21:36 +00:00
if (resource_int_value(device_get_name(dev), device_get_unit(dev),
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
"role", &tval) == 0 && tval != -1) {
tval &= (ISP_ROLE_INITIATOR|ISP_ROLE_TARGET);
isp->isp_role = tval;
device_printf(dev, "setting role to 0x%x\n", isp->isp_role);
Eliminate the use of the getenv_int stuff we'd been using (with a bitmap for selecting unit). Instead, use the resource hints mechanism. One unfortunate situation here is that there is no resource_quad_value function- which is what I needed for WWN boot time replacement. Worse- you can't store the hint as just plain hint.isp.0.nodewwn="0x50000000aaaa0001" because this gets interpreted as an int- incorrectly because it can't be converted to an int. I can't even get this as a string. To work around this particular case for nodewwn && portwwn setting, this rather grotesque form will be used: hint.isp.0.nodewwn="w50000000aaaa0001" hint.isp.0.portwwn="w50000000aaaa0002" At the same time, if we have no hinted WWN, set the default WWN (which, btw, gets overridden if the card has valid NVRAM, which is usual) to 0x400000007F000009ull (which translates to NAA == IPv4, 127.0.0.9). Eliminate more printf's and replace them either with device_printf or isp_prt calls.
2001-03-01 02:21:36 +00:00
} else {
Shuffle around how we do isp_disable management- make sure we return 0 so the unit number doesn't get reused. Make sure that if we've compiled for ISP_TARGET_MODE we set the default role to be ISP_ROLE_INITIATOR|ISP_ROLE_TARGET. Do some misc other cleanups.
2001-02-11 03:53:23 +00:00
#ifdef ISP_TARGET_MODE
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp->isp_role = ISP_ROLE_TARGET;
Shuffle around how we do isp_disable management- make sure we return 0 so the unit number doesn't get reused. Make sure that if we've compiled for ISP_TARGET_MODE we set the default role to be ISP_ROLE_INITIATOR|ISP_ROLE_TARGET. Do some misc other cleanups.
2001-02-11 03:53:23 +00:00
#else
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp->isp_role = ISP_DEFAULT_ROLES;
Shuffle around how we do isp_disable management- make sure we return 0 so the unit number doesn't get reused. Make sure that if we've compiled for ISP_TARGET_MODE we set the default role to be ISP_ROLE_INITIATOR|ISP_ROLE_TARGET. Do some misc other cleanups.
2001-02-11 03:53:23 +00:00
#endif
add in a boot environment isp_disable flag
1999-07-06 01:24:20 +00:00
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
tval = 0;
if (resource_int_value(device_get_name(dev), device_get_unit(dev),
"fwload_disable", &tval) == 0 && tval != 0) {
isp->isp_confopts |= ISP_CFG_NORELOAD;
}
tval = 0;
if (resource_int_value(device_get_name(dev), device_get_unit(dev),
"ignore_nvram", &tval) == 0 && tval != 0) {
isp->isp_confopts |= ISP_CFG_NONVRAM;
}
tval = 0;
if (resource_int_value(device_get_name(dev), device_get_unit(dev),
"fullduplex", &tval) == 0 && tval != 0) {
isp->isp_confopts |= ISP_CFG_FULL_DUPLEX;
}
#ifdef ISP_FW_CRASH_DUMP
tval = 0;
if (resource_int_value(device_get_name(dev), device_get_unit(dev),
"fw_dump_enable", &tval) == 0 && tval != 0) {
size_t amt = 0;
if (IS_2200(isp)) {
amt = QLA2200_RISC_IMAGE_DUMP_SIZE;
} else if (IS_23XX(isp)) {
amt = QLA2300_RISC_IMAGE_DUMP_SIZE;
}
if (amt) {
FCPARAM(isp)->isp_dump_data =
malloc(amt, M_DEVBUF, M_WAITOK | M_ZERO);
} else {
device_printf(dev,
"f/w crash dumps not supported for this model\n");
}
}
#endif
sptr = 0;
if (resource_string_value(device_get_name(dev), device_get_unit(dev),
"topology", (const char **) &sptr) == 0 && sptr != 0) {
if (strcmp(sptr, "lport") == 0) {
isp->isp_confopts |= ISP_CFG_LPORT;
} else if (strcmp(sptr, "nport") == 0) {
isp->isp_confopts |= ISP_CFG_NPORT;
} else if (strcmp(sptr, "lport-only") == 0) {
isp->isp_confopts |= ISP_CFG_LPORT_ONLY;
} else if (strcmp(sptr, "nport-only") == 0) {
isp->isp_confopts |= ISP_CFG_NPORT_ONLY;
}
}
/*
* Because the resource_*_value functions can neither return
* 64 bit integer values, nor can they be directly coerced
* to interpret the right hand side of the assignment as
* you want them to interpret it, we have to force WWN
* hint replacement to specify WWN strings with a leading
* 'w' (e..g w50000000aaaa0001). Sigh.
*/
sptr = 0;
tval = resource_string_value(device_get_name(dev), device_get_unit(dev),
"portwwn", (const char **) &sptr);
if (tval == 0 && sptr != 0 && *sptr++ == 'w') {
char *eptr = 0;
isp->isp_osinfo.default_port_wwn = strtouq(sptr, &eptr, 16);
if (eptr < sptr + 16 || isp->isp_osinfo.default_port_wwn == 0) {
device_printf(dev, "mangled portwwn hint '%s'\n", sptr);
isp->isp_osinfo.default_port_wwn = 0;
} else {
isp->isp_confopts |= ISP_CFG_OWNWWPN;
}
}
if (isp->isp_osinfo.default_port_wwn == 0) {
isp->isp_osinfo.default_port_wwn = 0x400000007F000009ull;
}
sptr = 0;
tval = resource_string_value(device_get_name(dev), device_get_unit(dev),
"nodewwn", (const char **) &sptr);
if (tval == 0 && sptr != 0 && *sptr++ == 'w') {
char *eptr = 0;
isp->isp_osinfo.default_node_wwn = strtouq(sptr, &eptr, 16);
if (eptr < sptr + 16 || isp->isp_osinfo.default_node_wwn == 0) {
device_printf(dev, "mangled nodewwn hint '%s'\n", sptr);
isp->isp_osinfo.default_node_wwn = 0;
} else {
isp->isp_confopts |= ISP_CFG_OWNWWNN;
}
}
if (isp->isp_osinfo.default_node_wwn == 0) {
isp->isp_osinfo.default_node_wwn = 0x400000007F000009ull;
}
isp->isp_osinfo.default_id = -1;
if (resource_int_value(device_get_name(dev), device_get_unit(dev),
"iid", &tval) == 0) {
isp->isp_osinfo.default_id = tval;
isp->isp_confopts |= ISP_CFG_OWNLOOPID;
}
if (isp->isp_osinfo.default_id == -1) {
if (IS_FC(isp)) {
isp->isp_osinfo.default_id = 109;
} else {
isp->isp_osinfo.default_id = 7;
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
/*
* Set up logging levels.
*/
tval = 0;
(void) resource_int_value(device_get_name(dev), device_get_unit(dev),
"debug", &tval);
if (tval) {
isp->isp_dblev = tval;
} else {
isp->isp_dblev = ISP_LOGWARN|ISP_LOGERR;
}
if (bootverbose) {
isp->isp_dblev |= ISP_LOGCONFIG|ISP_LOGINFO;
}
}
static void
isp_get_pci_options(device_t dev, int *m1, int *m2)
{
int tval;
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
/*
PAE support changes that included at least some minimal actual testing with a kernel that booted.
2004-09-23 05:25:22 +00:00
* Which we should try first - memory mapping or i/o mapping?
*
* We used to try memory first followed by i/o on alpha, otherwise
* the reverse, but we should just try memory first all the time now.
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
*/
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
*m1 = PCIM_CMD_MEMEN;
*m2 = PCIM_CMD_PORTEN;
Eliminate the use of the getenv_int stuff we'd been using (with a bitmap for selecting unit). Instead, use the resource hints mechanism. One unfortunate situation here is that there is no resource_quad_value function- which is what I needed for WWN boot time replacement. Worse- you can't store the hint as just plain hint.isp.0.nodewwn="0x50000000aaaa0001" because this gets interpreted as an int- incorrectly because it can't be converted to an int. I can't even get this as a string. To work around this particular case for nodewwn && portwwn setting, this rather grotesque form will be used: hint.isp.0.nodewwn="w50000000aaaa0001" hint.isp.0.portwwn="w50000000aaaa0002" At the same time, if we have no hinted WWN, set the default WWN (which, btw, gets overridden if the card has valid NVRAM, which is usual) to 0x400000007F000009ull (which translates to NAA == IPv4, 127.0.0.9). Eliminate more printf's and replace them either with device_printf or isp_prt calls.
2001-03-01 02:21:36 +00:00
tval = 0;
if (resource_int_value(device_get_name(dev), device_get_unit(dev),
"prefer_iomap", &tval) == 0 && tval != 0) {
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
*m1 = PCIM_CMD_PORTEN;
*m2 = PCIM_CMD_MEMEN;
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
}
Eliminate the use of the getenv_int stuff we'd been using (with a bitmap for selecting unit). Instead, use the resource hints mechanism. One unfortunate situation here is that there is no resource_quad_value function- which is what I needed for WWN boot time replacement. Worse- you can't store the hint as just plain hint.isp.0.nodewwn="0x50000000aaaa0001" because this gets interpreted as an int- incorrectly because it can't be converted to an int. I can't even get this as a string. To work around this particular case for nodewwn && portwwn setting, this rather grotesque form will be used: hint.isp.0.nodewwn="w50000000aaaa0001" hint.isp.0.portwwn="w50000000aaaa0002" At the same time, if we have no hinted WWN, set the default WWN (which, btw, gets overridden if the card has valid NVRAM, which is usual) to 0x400000007F000009ull (which translates to NAA == IPv4, 127.0.0.9). Eliminate more printf's and replace them either with device_printf or isp_prt calls.
2001-03-01 02:21:36 +00:00
tval = 0;
if (resource_int_value(device_get_name(dev), device_get_unit(dev),
"prefer_memmap", &tval) == 0 && tval != 0) {
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
*m1 = PCIM_CMD_MEMEN;
*m2 = PCIM_CMD_PORTEN;
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
}
#endif
static int
isp_pci_attach(device_t dev)
{
struct resource *regs, *irq;
int rtp, rgd, iqd, m1, m2;
uint32_t data, cmd, linesz, psize, basetype;
struct isp_pcisoftc *pcs;
ispsoftc_t *isp = NULL;
struct ispmdvec *mdvp;
#if __FreeBSD_version >= 500000
int locksetup = 0;
#endif
pcs = device_get_softc(dev);
if (pcs == NULL) {
device_printf(dev, "cannot get softc\n");
return (ENOMEM);
}
memset(pcs, 0, sizeof (*pcs));
pcs->pci_dev = dev;
isp = &pcs->pci_isp;
/*
* Get Generic Options
*/
isp_get_options(dev, isp);
/*
* Check to see if options have us disabled
*/
if (isp->isp_osinfo.disabled) {
/*
* But return zero to preserve unit numbering
*/
return (0);
}
/*
* Get PCI options- which in this case are just mapping preferences.
*/
isp_get_pci_options(dev, &m1, &m2);
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
Read the board revision and trim cache line size back from 16 to 1 for early revision 2100 boards. Make sure to turn ROM off for these boards.
1999-04-04 01:14:02 +00:00
linesz = PCI_DFLT_LNSZ;
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
irq = regs = NULL;
rgd = rtp = iqd = 0;
Fix some incorrectly swapped fields in an ICB. Access a PCI register with correct width. Obtained from: Dmitry Valeryevich Trikoz
2005-05-11 03:00:50 +00:00
cmd = pci_read_config(dev, PCIR_COMMAND, 2);
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (cmd & m1) {
rtp = (m1 == PCIM_CMD_MEMEN)? SYS_RES_MEMORY : SYS_RES_IOPORT;
rgd = (m1 == PCIM_CMD_MEMEN)? MEM_MAP_REG : IO_MAP_REG;
Convert callers to the new bus_alloc_resource_any(9) API. Submitted by: Mark Santcroos <marks@ripe.net> Reviewed by: imp, dfr, bde
2004-03-17 17:50:55 +00:00
regs = bus_alloc_resource_any(dev, rtp, &rgd, RF_ACTIVE);
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (regs == NULL && (cmd & m2)) {
rtp = (m2 == PCIM_CMD_MEMEN)? SYS_RES_MEMORY : SYS_RES_IOPORT;
rgd = (m2 == PCIM_CMD_MEMEN)? MEM_MAP_REG : IO_MAP_REG;
Convert callers to the new bus_alloc_resource_any(9) API. Submitted by: Mark Santcroos <marks@ripe.net> Reviewed by: imp, dfr, bde
2004-03-17 17:50:55 +00:00
regs = bus_alloc_resource_any(dev, rtp, &rgd, RF_ACTIVE);
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
}
if (regs == NULL) {
device_printf(dev, "unable to map any ports\n");
goto bad;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
if (bootverbose) {
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
device_printf(dev, "using %s space register mapping\n",
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
(rgd == IO_MAP_REG)? "I/O" : "Memory");
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
pcs->pci_dev = dev;
pcs->pci_reg = regs;
pcs->pci_st = rman_get_bustag(regs);
pcs->pci_sh = rman_get_bushandle(regs);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
pcs->pci_poff[BIU_BLOCK >> _BLK_REG_SHFT] = BIU_REGS_OFF;
pcs->pci_poff[MBOX_BLOCK >> _BLK_REG_SHFT] = PCI_MBOX_REGS_OFF;
pcs->pci_poff[SXP_BLOCK >> _BLK_REG_SHFT] = PCI_SXP_REGS_OFF;
pcs->pci_poff[RISC_BLOCK >> _BLK_REG_SHFT] = PCI_RISC_REGS_OFF;
pcs->pci_poff[DMA_BLOCK >> _BLK_REG_SHFT] = DMA_REGS_OFF;
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
mdvp = &mdvec;
basetype = ISP_HA_SCSI_UNKNOWN;
psize = sizeof (sdparam);
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP1020) {
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
mdvp = &mdvec;
basetype = ISP_HA_SCSI_UNKNOWN;
psize = sizeof (sdparam);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP1080) {
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
mdvp = &mdvec_1080;
basetype = ISP_HA_SCSI_1080;
psize = sizeof (sdparam);
pcs->pci_poff[DMA_BLOCK >> _BLK_REG_SHFT] =
ISP1080_DMA_REGS_OFF;
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP1240) {
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
mdvp = &mdvec_1080;
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
basetype = ISP_HA_SCSI_1240;
psize = 2 * sizeof (sdparam);
pcs->pci_poff[DMA_BLOCK >> _BLK_REG_SHFT] =
ISP1080_DMA_REGS_OFF;
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP1280) {
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
mdvp = &mdvec_1080;
basetype = ISP_HA_SCSI_1280;
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
psize = 2 * sizeof (sdparam);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
pcs->pci_poff[DMA_BLOCK >> _BLK_REG_SHFT] =
This should enable 10160 support. As best as I can tell, the same f/w as 12160 is used, and otherwise, this is just a single channel variant of the 10160. MFC after: 0 days
2002-10-11 17:28:01 +00:00
ISP1080_DMA_REGS_OFF;
}
if (pci_get_devid(dev) == PCI_QLOGIC_ISP10160) {
mdvp = &mdvec_12160;
basetype = ISP_HA_SCSI_10160;
psize = sizeof (sdparam);
pcs->pci_poff[DMA_BLOCK >> _BLK_REG_SHFT] =
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
ISP1080_DMA_REGS_OFF;
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP12160) {
mdvp = &mdvec_12160;
basetype = ISP_HA_SCSI_12160;
psize = 2 * sizeof (sdparam);
pcs->pci_poff[DMA_BLOCK >> _BLK_REG_SHFT] =
ISP1080_DMA_REGS_OFF;
}
if (pci_get_devid(dev) == PCI_QLOGIC_ISP2100) {
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
mdvp = &mdvec_2100;
basetype = ISP_HA_FC_2100;
psize = sizeof (fcparam);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
pcs->pci_poff[MBOX_BLOCK >> _BLK_REG_SHFT] =
PCI_MBOX_REGS2100_OFF;
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (pci_get_revid(dev) < 3) {
Read the board revision and trim cache line size back from 16 to 1 for early revision 2100 boards. Make sure to turn ROM off for these boards.
1999-04-04 01:14:02 +00:00
/*
* XXX: Need to get the actual revision
* XXX: number of the 2100 FB. At any rate,
* XXX: lower cache line size for early revision
* XXX; boards.
*/
linesz = 1;
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP2200) {
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
mdvp = &mdvec_2200;
basetype = ISP_HA_FC_2200;
psize = sizeof (fcparam);
pcs->pci_poff[MBOX_BLOCK >> _BLK_REG_SHFT] =
PCI_MBOX_REGS2100_OFF;
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP2300) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
mdvp = &mdvec_2300;
basetype = ISP_HA_FC_2300;
psize = sizeof (fcparam);
pcs->pci_poff[MBOX_BLOCK >> _BLK_REG_SHFT] =
PCI_MBOX_REGS2300_OFF;
}
Support the DELL OEM 2312 cards (1077,6312). Many thanks to Stormweb for making the h/w available for testing. MFC after: 2 days
2005-01-23 06:23:55 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP2312 ||
pci_get_devid(dev) == PCI_QLOGIC_ISP6312) {
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
mdvp = &mdvec_2300;
basetype = ISP_HA_FC_2312;
psize = sizeof (fcparam);
pcs->pci_poff[MBOX_BLOCK >> _BLK_REG_SHFT] =
PCI_MBOX_REGS2300_OFF;
}
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP2322 ||
pci_get_devid(dev) == PCI_QLOGIC_ISP6322) {
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
mdvp = &mdvec_2300;
basetype = ISP_HA_FC_2322;
psize = sizeof (fcparam);
pcs->pci_poff[MBOX_BLOCK >> _BLK_REG_SHFT] =
PCI_MBOX_REGS2300_OFF;
}
Put in at least an attempt to ID the 2422 (4Gb part)
2006-01-26 05:04:35 +00:00
if (pci_get_devid(dev) == PCI_QLOGIC_ISP2422) {
mdvp = &mdvec_2300;
basetype = ISP_HA_FC_2422;
psize = sizeof (fcparam);
pcs->pci_poff[MBOX_BLOCK >> _BLK_REG_SHFT] =
PCI_MBOX_REGS2300_OFF;
}
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
isp = &pcs->pci_isp;
Convert more malloc+bzero to malloc+M_ZERO. Submitted by: josh@zipperup.org Submitted by: Robert Drehmel <robd@gmx.net>
2000-12-08 21:51:06 +00:00
isp->isp_param = malloc(psize, M_DEVBUF, M_NOWAIT | M_ZERO);
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
if (isp->isp_param == NULL) {
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_printf(dev, "cannot allocate parameter data\n");
goto bad;
Clean up 2.2.X support (which might have to be cleaned up again after some of the previous commits). Add in support for the 1240 dual channel ISP card. Try the dance of unmapping a PCI interrupt if we don't configure (if that ever works it'll be helpful).
1999-05-11 04:53:57 +00:00
}
isp->isp_mdvec = mdvp;
isp->isp_type = basetype;
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
isp->isp_revision = pci_get_revid(dev);
Eliminate the use of the getenv_int stuff we'd been using (with a bitmap for selecting unit). Instead, use the resource hints mechanism. One unfortunate situation here is that there is no resource_quad_value function- which is what I needed for WWN boot time replacement. Worse- you can't store the hint as just plain hint.isp.0.nodewwn="0x50000000aaaa0001" because this gets interpreted as an int- incorrectly because it can't be converted to an int. I can't even get this as a string. To work around this particular case for nodewwn && portwwn setting, this rather grotesque form will be used: hint.isp.0.nodewwn="w50000000aaaa0001" hint.isp.0.portwwn="w50000000aaaa0002" At the same time, if we have no hinted WWN, set the default WWN (which, btw, gets overridden if the card has valid NVRAM, which is usual) to 0x400000007F000009ull (which translates to NAA == IPv4, 127.0.0.9). Eliminate more printf's and replace them either with device_printf or isp_prt calls.
2001-03-01 02:21:36 +00:00
isp->isp_dev = dev;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
#if __FreeBSD_version >= 700000
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
/*
* Try and find firmware for this device.
*/
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
{
char fwname[32];
unsigned int did = pci_get_device(dev);
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
/*
* Map a few pci ids to fw names
*/
switch (did) {
case PCI_PRODUCT_QLOGIC_ISP1020:
did = 0x1040;
break;
case PCI_PRODUCT_QLOGIC_ISP1240:
did = 0x1080;
break;
case PCI_PRODUCT_QLOGIC_ISP10160:
case PCI_PRODUCT_QLOGIC_ISP12160:
did = 0x12160;
break;
case PCI_PRODUCT_QLOGIC_ISP6312:
case PCI_PRODUCT_QLOGIC_ISP2312:
did = 0x2300;
break;
case PCI_PRODUCT_QLOGIC_ISP6322:
did = 0x2322;
break;
default:
break;
}
isp->isp_osinfo.fw = NULL;
if (isp->isp_role & ISP_ROLE_TARGET) {
snprintf(fwname, sizeof (fwname), "isp_%04x_it", did);
isp->isp_osinfo.fw = firmware_get(fwname);
}
if (isp->isp_osinfo.fw == NULL) {
snprintf(fwname, sizeof (fwname), "isp_%04x", did);
isp->isp_osinfo.fw = firmware_get(fwname);
}
if (isp->isp_osinfo.fw != NULL) {
union {
const void *fred;
uint16_t *bob;
} u;
u.fred = isp->isp_osinfo.fw->data;
isp->isp_mdvec->dv_ispfw = u.bob;
}
}
#else
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (isp_get_firmware_p) {
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
int device = (int) pci_get_device(dev);
#ifdef ISP_TARGET_MODE
(*isp_get_firmware_p)(0, 1, device, &mdvp->dv_ispfw);
#else
(*isp_get_firmware_p)(0, 0, device, &mdvp->dv_ispfw);
#endif
}
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
#endif
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
/*
* Make sure that SERR, PERR, WRITE INVALIDATE and BUSMASTER
* are set.
*/
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
cmd |= PCIM_CMD_SEREN | PCIM_CMD_PERRESPEN |
PCIM_CMD_BUSMASTEREN | PCIM_CMD_INVEN;
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
if (IS_2300(isp)) { /* per QLogic errata */
cmd &= ~PCIM_CMD_INVEN;
}
Convert isp(4) and ispfw(4) to use firmware(9) to manage firmware loading for the QLogic cards. Because isp(4) exists before the root is mounted, it's not really possible for us to use the kernel's linker to load modules directly from disk- that's really too bad. However, the this is still a net win in in that the firmware has been split up on a per chip (and in some cases, functionality) basis, so the amount of stuff loaded *can* be substantially less than the 1.5MB of firmware images that ispfw now manages. That is, each specific f/w set is now also built as a module. For example, QLogic 2322 f/w is built as isp_2322.ko and Initiator/Target 1080 firmware is built as isp_1080_it.ko. For compatibility purposes (i.e., to perturb folks the least), we also still build all of the firmware as one ispfw.ko module. This allows us to let 'ispfw_LOAD' keep on working in existing loader.conf files. If you now want to strip this down to just the firmware for your h/w, you can then change loader.conf to load the f/w you specifically want. We also still allow for ispfw to be statically built (e.g., for PAE and sparc64). Future changes will look at f/w unloading and also role switching that then uses the kernel linker to load different ips f/w sets. MFC after: 2 months
2006-07-09 17:50:20 +00:00
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
if (IS_23XX(isp)) {
/*
* Can't tell if ROM will hang on 'ABOUT FIRMWARE' command.
*/
isp->isp_touched = 1;
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (IS_2322(isp) || pci_get_devid(dev) == PCI_QLOGIC_ISP6312) {
cmd &= ~PCIM_CMD_INTX_DISABLE;
}
Fix some incorrectly swapped fields in an ICB. Access a PCI register with correct width. Obtained from: Dmitry Valeryevich Trikoz
2005-05-11 03:00:50 +00:00
pci_write_config(dev, PCIR_COMMAND, cmd, 2);
Read the board revision and trim cache line size back from 16 to 1 for early revision 2100 boards. Make sure to turn ROM off for these boards.
1999-04-04 01:14:02 +00:00
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
/*
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
* Make sure the Cache Line Size register is set sensibly.
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
*/
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
data = pci_read_config(dev, PCIR_CACHELNSZ, 1);
Read the board revision and trim cache line size back from 16 to 1 for early revision 2100 boards. Make sure to turn ROM off for these boards.
1999-04-04 01:14:02 +00:00
if (data != linesz) {
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
data = PCI_DFLT_LNSZ;
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
isp_prt(isp, ISP_LOGCONFIG, "set PCI line size to %d", data);
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
pci_write_config(dev, PCIR_CACHELNSZ, data, 1);
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
}
Read the board revision and trim cache line size back from 16 to 1 for early revision 2100 boards. Make sure to turn ROM off for these boards.
1999-04-04 01:14:02 +00:00
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
/*
* Make sure the Latency Timer is sane.
*/
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
data = pci_read_config(dev, PCIR_LATTIMER, 1);
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
if (data < PCI_DFLT_LTNCY) {
data = PCI_DFLT_LTNCY;
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
isp_prt(isp, ISP_LOGCONFIG, "set PCI latency to %d", data);
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
pci_write_config(dev, PCIR_LATTIMER, data, 1);
Cleanup. Set all PCI parameters of importance. Set a define that will allow us via config options prefer mem space to I/O space.
1999-02-09 01:12:52 +00:00
}
Read the board revision and trim cache line size back from 16 to 1 for early revision 2100 boards. Make sure to turn ROM off for these boards.
1999-04-04 01:14:02 +00:00
/*
* Make sure we've disabled the ROM.
*/
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
data = pci_read_config(dev, PCIR_ROMADDR, 4);
Read the board revision and trim cache line size back from 16 to 1 for early revision 2100 boards. Make sure to turn ROM off for these boards.
1999-04-04 01:14:02 +00:00
data &= ~1;
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
pci_write_config(dev, PCIR_ROMADDR, data, 4);
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
iqd = 0;
Convert callers to the new bus_alloc_resource_any(9) API. Submitted by: Mark Santcroos <marks@ripe.net> Reviewed by: imp, dfr, bde
2004-03-17 17:50:55 +00:00
irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &iqd,
RF_ACTIVE | RF_SHAREABLE);
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (irq == NULL) {
device_printf(dev, "could not allocate interrupt\n");
goto bad;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#if __FreeBSD_version >= 500000
Set up to do a local interrupt fielding before calling common code- allows us to grab lock as we should.
2000-12-29 19:10:16 +00:00
/* Make sure the lock is set up. */
Change callers of mtx_init() to pass in an appropriate lock type name. In most cases NULL is passed, but in some cases such as network driver locks (which use the MTX_NETWORK_LOCK macro) and UMA zone locks, a name is used. Tested on: i386, alpha, sparc64
2002-04-04 21:03:38 +00:00
mtx_init(&isp->isp_osinfo.lock, "isp", NULL, MTX_DEF);
Set up to do a local interrupt fielding before calling common code- allows us to grab lock as we should.
2000-12-29 19:10:16 +00:00
locksetup++;
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#endif
Set up to do a local interrupt fielding before calling common code- allows us to grab lock as we should.
2000-12-29 19:10:16 +00:00
Make sure that if are in fact using 'full SMP', make the interrupt flags include INTR_MPSAFE. Put the flags in a common place so that both isp_sbus && isp_pci DTRT. In isp_mbxdma setup, drop any locks prior to calling things like bus_dmatag_create. This gets rid of these obnoxious WITNESS messages about 'sleeping with locks held' blah blah blah blah blah.
2002-07-25 16:02:09 +00:00
if (bus_setup_intr(dev, irq, ISP_IFLAGS, isp_pci_intr, isp, &pcs->ih)) {
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
device_printf(dev, "could not setup interrupt\n");
goto bad;
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
/*
* Last minute checks...
*/
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_23XX(isp)) {
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
isp->isp_port = pci_get_function(dev);
}
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
/*
* Make sure we're in reset state.
*/
Whoops! Forgot to commit this when I committed the other (turnin on locks) change. Sorry about that.
2000-10-25 04:40:49 +00:00
ISP_LOCK(isp);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
isp_reset(isp);
if (isp->isp_state != ISP_RESETSTATE) {
Whoops! Forgot to commit this when I committed the other (turnin on locks) change. Sorry about that.
2000-10-25 04:40:49 +00:00
ISP_UNLOCK(isp);
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
goto bad;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
isp_init(isp);
Shuffle around how we do isp_disable management- make sure we return 0 so the unit number doesn't get reused. Make sure that if we've compiled for ISP_TARGET_MODE we set the default role to be ISP_ROLE_INITIATOR|ISP_ROLE_TARGET. Do some misc other cleanups.
2001-02-11 03:53:23 +00:00
if (isp->isp_role != ISP_ROLE_NONE && isp->isp_state != ISP_INITSTATE) {
isp_uninit(isp);
ISP_UNLOCK(isp);
goto bad;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
isp_attach(isp);
Shuffle around how we do isp_disable management- make sure we return 0 so the unit number doesn't get reused. Make sure that if we've compiled for ISP_TARGET_MODE we set the default role to be ISP_ROLE_INITIATOR|ISP_ROLE_TARGET. Do some misc other cleanups.
2001-02-11 03:53:23 +00:00
if (isp->isp_role != ISP_ROLE_NONE && isp->isp_state != ISP_RUNSTATE) {
isp_uninit(isp);
ISP_UNLOCK(isp);
goto bad;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
/*
* XXXX: Here is where we might unload the f/w module
* XXXX: (or decrease the reference count to it).
*/
Whoops! Forgot to commit this when I committed the other (turnin on locks) change. Sorry about that.
2000-10-25 04:40:49 +00:00
ISP_UNLOCK(isp);
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
return (0);
bad:
if (pcs && pcs->ih) {
(void) bus_teardown_intr(dev, irq, pcs->ih);
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#if __FreeBSD_version >= 500000
Whoops! Forgot to commit this when I committed the other (turnin on locks) change. Sorry about that.
2000-10-25 04:40:49 +00:00
if (locksetup && isp) {
mtx_destroy(&isp->isp_osinfo.lock);
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#endif
Whoops! Forgot to commit this when I committed the other (turnin on locks) change. Sorry about that.
2000-10-25 04:40:49 +00:00
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (irq) {
(void) bus_release_resource(dev, SYS_RES_IRQ, iqd, irq);
}
Whoops! Forgot to commit this when I committed the other (turnin on locks) change. Sorry about that.
2000-10-25 04:40:49 +00:00
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (regs) {
(void) bus_release_resource(dev, rtp, rgd, regs);
}
Whoops! Forgot to commit this when I committed the other (turnin on locks) change. Sorry about that.
2000-10-25 04:40:49 +00:00
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
if (pcs) {
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
if (pcs->pci_isp.isp_param) {
#ifdef ISP_FW_CRASH_DUMP
if (IS_FC(isp) && FCPARAM(isp)->isp_dump_data) {
free(FCPARAM(isp)->isp_dump_data, M_DEVBUF);
}
#endif
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
free(pcs->pci_isp.isp_param, M_DEVBUF);
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
}
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
}
Whoops! Forgot to commit this when I committed the other (turnin on locks) change. Sorry about that.
2000-10-25 04:40:49 +00:00
Clean up firmware load issues and remove darn near all config options. Force alphas to prefer mem mapping as the default. Basically, we have a pointer to a function which we can call which will return us a pointer to firmware for the card we have. We call this function (if it's non-NULL) with the address of our mdvec f/w pointer. The way this works is that if ispfw (as a module or a static) is loaded, it initializes the pointer in isp_pci, so we can call into to it to fetch a pointer to a f/w set. If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers, which then get zeroed out so we don't have any references to data that's now gone from kernel memory. Removing the f/w saves ~360KBytes. Alas, there is no autounload mechanism that works for is here.
2000-06-18 05:18:55 +00:00
/*
* XXXX: Here is where we might unload the f/w module
* XXXX: (or decrease the reference count to it).
*/
Add in 12160 (Ultra3) support. Redo things to use the newbus code. Approved: jkh@freebsd.org PR: 16141
2000-02-11 19:45:17 +00:00
return (ENXIO);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Set up to do a local interrupt fielding before calling common code- allows us to grab lock as we should.
2000-12-29 19:10:16 +00:00
static void
isp_pci_intr(void *arg)
{
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t *isp = arg;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t isr, sema, mbox;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
Set up to do a local interrupt fielding before calling common code- allows us to grab lock as we should.
2000-12-29 19:10:16 +00:00
ISP_LOCK(isp);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
isp->isp_intcnt++;
if (ISP_READ_ISR(isp, &isr, &sema, &mbox) == 0) {
isp->isp_intbogus++;
} else {
If we're on an interrupt stack, mark things so that we don't try and cv_wait for mailbox commands to complete if we start them from here. Fix residuals for target mode such that we only check the residual and set it in the CTIO if this is the last CTIO (when we're sending status). MFC after: 4 weeks
2001-09-04 21:45:57 +00:00
int iok = isp->isp_osinfo.intsok;
isp->isp_osinfo.intsok = 0;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
isp_intr(isp, isr, sema, mbox);
If we're on an interrupt stack, mark things so that we don't try and cv_wait for mailbox commands to complete if we start them from here. Fix residuals for target mode such that we only check the residual and set it in the CTIO if this is the last CTIO (when we're sending status). MFC after: 4 weeks
2001-09-04 21:45:57 +00:00
isp->isp_osinfo.intsok = iok;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
}
Set up to do a local interrupt fielding before calling common code- allows us to grab lock as we should.
2000-12-29 19:10:16 +00:00
ISP_UNLOCK(isp);
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
#define IspVirt2Off(a, x) \
(((struct isp_pcisoftc *)a)->pci_poff[((x) & _BLK_REG_MASK) >> \
The register offset is within 4K, not 256 bytes, for some QLogic cards.
2006-08-14 05:36:26 +00:00
_BLK_REG_SHFT] + ((x) & 0xfff))
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
#define BXR2(pcs, off) \
bus_space_read_2(pcs->pci_st, pcs->pci_sh, off)
#define BXW2(pcs, off, v) \
bus_space_write_2(pcs->pci_st, pcs->pci_sh, off, v)
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
static __inline int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_rd_debounced(ispsoftc_t *isp, int off, uint16_t *rp)
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
{
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *) isp;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t val0, val1;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
int i = 0;
do {
val0 = BXR2(pcs, IspVirt2Off(isp, off));
val1 = BXR2(pcs, IspVirt2Off(isp, off));
} while (val0 != val1 && ++i < 1000);
if (val0 != val1) {
return (1);
}
*rp = val0;
return (0);
}
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_rd_isr(ispsoftc_t *isp, uint16_t *isrp,
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t *semap, uint16_t *mbp)
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
{
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *) isp;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t isr, sema;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
if (IS_2100(isp)) {
if (isp_pci_rd_debounced(isp, BIU_ISR, &isr)) {
return (0);
}
if (isp_pci_rd_debounced(isp, BIU_SEMA, &sema)) {
return (0);
}
} else {
isr = BXR2(pcs, IspVirt2Off(isp, BIU_ISR));
sema = BXR2(pcs, IspVirt2Off(isp, BIU_SEMA));
}
isp_prt(isp, ISP_LOGDEBUG3, "ISR 0x%x SEMA 0x%x", isr, sema);
isr &= INT_PENDING_MASK(isp);
sema &= BIU_SEMA_LOCK;
if (isr == 0 && sema == 0) {
return (0);
}
*isrp = isr;
if ((*semap = sema) != 0) {
if (IS_2100(isp)) {
if (isp_pci_rd_debounced(isp, OUTMAILBOX0, mbp)) {
return (0);
}
} else {
*mbp = BXR2(pcs, IspVirt2Off(isp, OUTMAILBOX0));
}
}
return (1);
}
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_rd_isr_2300(ispsoftc_t *isp, uint16_t *isrp,
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t *semap, uint16_t *mbox0p)
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
{
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *) isp;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
uint16_t hccr;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint32_t r2hisr;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
Whups- remember to zero the isr pointer arg.
2001-10-06 19:34:43 +00:00
if (!(BXR2(pcs, IspVirt2Off(isp, BIU_ISR) & BIU2100_ISR_RISC_INT))) {
*isrp = 0;
Respect QLogic's errata- read BIU_ISR even on the 2300 to see if there's an interrupt (avoids PCI parity errors which can occur on the 2312 if you access some registers from the host at the same time the RISC on the 2312 is C accessing them). MFC after: 1 day
2001-10-06 19:19:24 +00:00
return (0);
Whups- remember to zero the isr pointer arg.
2001-10-06 19:34:43 +00:00
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
r2hisr = bus_space_read_4(pcs->pci_st, pcs->pci_sh,
IspVirt2Off(pcs, BIU_R2HSTSLO));
isp_prt(isp, ISP_LOGDEBUG3, "RISC2HOST ISR 0x%x", r2hisr);
if ((r2hisr & BIU_R2HST_INTR) == 0) {
*isrp = 0;
return (0);
}
switch (r2hisr & BIU_R2HST_ISTAT_MASK) {
case ISPR2HST_ROM_MBX_OK:
case ISPR2HST_ROM_MBX_FAIL:
case ISPR2HST_MBX_OK:
case ISPR2HST_MBX_FAIL:
case ISPR2HST_ASYNC_EVENT:
Fix the incorrect parsing of the Risc2Host isr. For RIO_16, fast posting command completion, and fast post CTIO completion, the upper half of Risc2Host is a copy of mailbox #1- *not* mailbox #0. MFC after: 1 day
2002-08-17 17:21:54 +00:00
*isrp = r2hisr & 0xffff;
*mbox0p = (r2hisr >> 16);
*semap = 1;
return (1);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
case ISPR2HST_RIO_16:
Fix the incorrect parsing of the Risc2Host isr. For RIO_16, fast posting command completion, and fast post CTIO completion, the upper half of Risc2Host is a copy of mailbox #1- *not* mailbox #0. MFC after: 1 day
2002-08-17 17:21:54 +00:00
*isrp = r2hisr & 0xffff;
*mbox0p = ASYNC_RIO1;
*semap = 1;
return (1);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
case ISPR2HST_FPOST:
Fix the incorrect parsing of the Risc2Host isr. For RIO_16, fast posting command completion, and fast post CTIO completion, the upper half of Risc2Host is a copy of mailbox #1- *not* mailbox #0. MFC after: 1 day
2002-08-17 17:21:54 +00:00
*isrp = r2hisr & 0xffff;
*mbox0p = ASYNC_CMD_CMPLT;
*semap = 1;
return (1);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
case ISPR2HST_FPOST_CTIO:
*isrp = r2hisr & 0xffff;
Fix the incorrect parsing of the Risc2Host isr. For RIO_16, fast posting command completion, and fast post CTIO completion, the upper half of Risc2Host is a copy of mailbox #1- *not* mailbox #0. MFC after: 1 day
2002-08-17 17:21:54 +00:00
*mbox0p = ASYNC_CTIO_DONE;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
*semap = 1;
return (1);
case ISPR2HST_RSPQ_UPDATE:
*isrp = r2hisr & 0xffff;
*mbox0p = 0;
*semap = 0;
return (1);
default:
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
hccr = ISP_READ(isp, HCCR);
if (hccr & HCCR_PAUSE) {
ISP_WRITE(isp, HCCR, HCCR_RESET);
isp_prt(isp, ISP_LOGERR,
"RISC paused at interrupt (%x->%x\n", hccr,
ISP_READ(isp, HCCR));
} else {
isp_prt(isp, ISP_LOGERR, "unknown interrerupt 0x%x\n",
r2hisr);
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
return (0);
}
}
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
static uint16_t
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_rd_reg(ispsoftc_t *isp, int regoff)
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
{
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t rv;
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *) isp;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
int oldconf = 0;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
if ((regoff & _BLK_REG_MASK) == SXP_BLOCK) {
/*
* We will assume that someone has paused the RISC processor.
*/
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
oldconf = BXR2(pcs, IspVirt2Off(isp, BIU_CONF1));
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1),
oldconf | BIU_PCI_CONF1_SXP);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
rv = BXR2(pcs, IspVirt2Off(isp, regoff));
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
if ((regoff & _BLK_REG_MASK) == SXP_BLOCK) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1), oldconf);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
}
return (rv);
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_wr_reg(ispsoftc_t *isp, int regoff, uint16_t val)
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
{
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *) isp;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
int oldconf = 0;
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
if ((regoff & _BLK_REG_MASK) == SXP_BLOCK) {
/*
* We will assume that someone has paused the RISC processor.
*/
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
oldconf = BXR2(pcs, IspVirt2Off(isp, BIU_CONF1));
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1),
oldconf | BIU_PCI_CONF1_SXP);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
BXW2(pcs, IspVirt2Off(isp, regoff), val);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
if ((regoff & _BLK_REG_MASK) == SXP_BLOCK) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1), oldconf);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
}
}
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
static uint16_t
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_rd_reg_1080(ispsoftc_t *isp, int regoff)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
{
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t rv, oc = 0;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *) isp;
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
if ((regoff & _BLK_REG_MASK) == SXP_BLOCK ||
(regoff & _BLK_REG_MASK) == (SXP_BLOCK|SXP_BANK1_SELECT)) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t tc;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
/*
* We will assume that someone has paused the RISC processor.
*/
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
oc = BXR2(pcs, IspVirt2Off(isp, BIU_CONF1));
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
tc = oc & ~BIU_PCI1080_CONF1_DMA;
if (regoff & SXP_BANK1_SELECT)
tc |= BIU_PCI1080_CONF1_SXP1;
else
tc |= BIU_PCI1080_CONF1_SXP0;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1), tc);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
} else if ((regoff & _BLK_REG_MASK) == DMA_BLOCK) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
oc = BXR2(pcs, IspVirt2Off(isp, BIU_CONF1));
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1),
oc | BIU_PCI1080_CONF1_DMA);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
rv = BXR2(pcs, IspVirt2Off(isp, regoff));
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
if (oc) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1), oc);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
return (rv);
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_wr_reg_1080(ispsoftc_t *isp, int regoff, uint16_t val)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
{
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *) isp;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
int oc = 0;
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
if ((regoff & _BLK_REG_MASK) == SXP_BLOCK ||
(regoff & _BLK_REG_MASK) == (SXP_BLOCK|SXP_BANK1_SELECT)) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t tc;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
/*
* We will assume that someone has paused the RISC processor.
*/
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
oc = BXR2(pcs, IspVirt2Off(isp, BIU_CONF1));
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
tc = oc & ~BIU_PCI1080_CONF1_DMA;
if (regoff & SXP_BANK1_SELECT)
tc |= BIU_PCI1080_CONF1_SXP1;
else
tc |= BIU_PCI1080_CONF1_SXP0;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1), tc);
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
} else if ((regoff & _BLK_REG_MASK) == DMA_BLOCK) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
oc = BXR2(pcs, IspVirt2Off(isp, BIU_CONF1));
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1),
oc | BIU_PCI1080_CONF1_DMA);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
BXW2(pcs, IspVirt2Off(isp, regoff), val);
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
if (oc) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
BXW2(pcs, IspVirt2Off(isp, BIU_CONF1), oc);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
}
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
struct imush {
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t *isp;
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
int error;
};
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
static void imc(void *, bus_dma_segment_t *, int, int);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
static void
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
imc(void *arg, bus_dma_segment_t *segs, int nseg, int error)
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
{
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
struct imush *imushp = (struct imush *) arg;
if (error) {
imushp->error = error;
} else {
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t *isp =imushp->isp;
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_addr_t addr = segs->ds_addr;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
isp->isp_rquest_dma = addr;
addr += ISP_QUEUE_SIZE(RQUEST_QUEUE_LEN(isp));
isp->isp_result_dma = addr;
if (IS_FC(isp)) {
addr += ISP_QUEUE_SIZE(RESULT_QUEUE_LEN(isp));
FCPARAM(isp)->isp_scdma = addr;
}
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
}
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
/*
* Should be BUS_SPACE_MAXSIZE, but MAXPHYS is larger than BUS_SPACE_MAXSIZE
*/
#define ISP_NSEGS ((MAXPHYS / PAGE_SIZE) + 1)
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#if __FreeBSD_version < 500000
#define isp_dma_tag_create bus_dma_tag_create
#else
#define isp_dma_tag_create(a, b, c, d, e, f, g, h, i, j, k, z) \
bus_dma_tag_create(a, b, c, d, e, f, g, h, i, j, k, \
busdma_lock_mutex, &Giant, z)
#endif
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_mbxdma(ispsoftc_t *isp)
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
{
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *)isp;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
caddr_t base;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint32_t len;
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
int i, error, ns;
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
bus_size_t slim; /* segment size */
i386/PAE defines bus_size_t to be 32-bits when it likely should be 64-bits. Fixing it is left for another day, so just hack around it for now.
2006-02-04 03:41:48 +00:00
bus_addr_t llim; /* low limit of unavailable dma */
Actually, no, I had it wrong in 1.109. The arguments to bus_dma_create_tag are bus_addr_t, not bus_size_t. In any case, turn off DAC support entirely until it is revamped to actually work *correctly* for 64 bit platforms (not using a PAE definition and for both initiator and target mode).
2006-02-04 08:39:02 +00:00
bus_addr_t hlim; /* high limit of unavailable dma */
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
struct imush im;
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
/*
* Already been here? If so, leave...
*/
if (isp->isp_rquest) {
return (0);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
hlim = BUS_SPACE_MAXADDR;
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
if (IS_ULTRA2(isp) || IS_FC(isp) || IS_1240(isp)) {
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
slim = (bus_size_t) (1ULL << 32);
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
llim = BUS_SPACE_MAXADDR;
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
} else {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
llim = BUS_SPACE_MAXADDR_32BIT;
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
slim = (1 << 24);
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
}
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
/*
* XXX: We don't really support 64 bit target mode for parallel scsi yet
*/
#ifdef ISP_TARGET_MODE
if (IS_SCSI(isp) && sizeof (bus_addr_t) > 4) {
isp_prt(isp, ISP_LOGERR, "we cannot do DAC for SPI cards yet");
return (1);
}
#endif
Make sure that if are in fact using 'full SMP', make the interrupt flags include INTR_MPSAFE. Put the flags in a common place so that both isp_sbus && isp_pci DTRT. In isp_mbxdma setup, drop any locks prior to calling things like bus_dmatag_create. This gets rid of these obnoxious WITNESS messages about 'sleeping with locks held' blah blah blah blah blah.
2002-07-25 16:02:09 +00:00
ISP_UNLOCK(isp);
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
if (isp_dma_tag_create(NULL, 1, slim, llim, hlim,
NULL, NULL, BUS_SPACE_MAXSIZE, ISP_NSEGS, slim, 0, &pcs->dmat)) {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
isp_prt(isp, ISP_LOGERR, "could not create master dma tag");
Make sure that if are in fact using 'full SMP', make the interrupt flags include INTR_MPSAFE. Put the flags in a common place so that both isp_sbus && isp_pci DTRT. In isp_mbxdma setup, drop any locks prior to calling things like bus_dmatag_create. This gets rid of these obnoxious WITNESS messages about 'sleeping with locks held' blah blah blah blah blah.
2002-07-25 16:02:09 +00:00
ISP_LOCK(isp);
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
return (1);
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
}
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
len = sizeof (XS_T **) * isp->isp_maxcmds;
Back out M_* changes, per decision of the TRB. Approved by: trb
2003-02-19 05:47:46 +00:00
isp->isp_xflist = (XS_T **) malloc(len, M_DEVBUF, M_WAITOK | M_ZERO);
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
if (isp->isp_xflist == NULL) {
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
isp_prt(isp, ISP_LOGERR, "cannot alloc xflist array");
Make sure that if are in fact using 'full SMP', make the interrupt flags include INTR_MPSAFE. Put the flags in a common place so that both isp_sbus && isp_pci DTRT. In isp_mbxdma setup, drop any locks prior to calling things like bus_dmatag_create. This gets rid of these obnoxious WITNESS messages about 'sleeping with locks held' blah blah blah blah blah.
2002-07-25 16:02:09 +00:00
ISP_LOCK(isp);
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
return (1);
}
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
#ifdef ISP_TARGET_MODE
len = sizeof (void **) * isp->isp_maxcmds;
isp->isp_tgtlist = (void **) malloc(len, M_DEVBUF, M_WAITOK | M_ZERO);
if (isp->isp_tgtlist == NULL) {
isp_prt(isp, ISP_LOGERR, "cannot alloc tgtlist array");
ISP_LOCK(isp);
return (1);
}
#endif
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
len = sizeof (bus_dmamap_t) * isp->isp_maxcmds;
Back out M_* changes, per decision of the TRB. Approved by: trb
2003-02-19 05:47:46 +00:00
pcs->dmaps = (bus_dmamap_t *) malloc(len, M_DEVBUF, M_WAITOK);
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
if (pcs->dmaps == NULL) {
isp_prt(isp, ISP_LOGERR, "can't alloc dma map storage");
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
free(isp->isp_xflist, M_DEVBUF);
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
#ifdef ISP_TARGET_MODE
free(isp->isp_tgtlist, M_DEVBUF);
#endif
Make sure that if are in fact using 'full SMP', make the interrupt flags include INTR_MPSAFE. Put the flags in a common place so that both isp_sbus && isp_pci DTRT. In isp_mbxdma setup, drop any locks prior to calling things like bus_dmatag_create. This gets rid of these obnoxious WITNESS messages about 'sleeping with locks held' blah blah blah blah blah.
2002-07-25 16:02:09 +00:00
ISP_LOCK(isp);
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
return (1);
}
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
/*
* Allocate and map the request, result queues, plus FC scratch area.
*/
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
len = ISP_QUEUE_SIZE(RQUEST_QUEUE_LEN(isp));
len += ISP_QUEUE_SIZE(RESULT_QUEUE_LEN(isp));
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
if (IS_FC(isp)) {
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
len += ISP2100_SCRLEN;
}
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
ns = (len / PAGE_SIZE) + 1;
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
/*
* Create a tag for the control spaces- force it to within 32 bits.
*/
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
if (isp_dma_tag_create(pcs->dmat, QENTRY_LEN, slim,
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
NULL, NULL, len, ns, slim, 0, &isp->isp_cdmat)) {
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
isp_prt(isp, ISP_LOGERR,
"cannot create a dma tag for control spaces");
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
free(pcs->dmaps, M_DEVBUF);
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
free(isp->isp_xflist, M_DEVBUF);
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
#ifdef ISP_TARGET_MODE
free(isp->isp_tgtlist, M_DEVBUF);
#endif
Make sure that if are in fact using 'full SMP', make the interrupt flags include INTR_MPSAFE. Put the flags in a common place so that both isp_sbus && isp_pci DTRT. In isp_mbxdma setup, drop any locks prior to calling things like bus_dmatag_create. This gets rid of these obnoxious WITNESS messages about 'sleeping with locks held' blah blah blah blah blah.
2002-07-25 16:02:09 +00:00
ISP_LOCK(isp);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
return (1);
}
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
if (bus_dmamem_alloc(isp->isp_cdmat, (void **)&base, BUS_DMA_NOWAIT,
&isp->isp_cdmap) != 0) {
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
isp_prt(isp, ISP_LOGERR,
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
"cannot allocate %d bytes of CCB memory", len);
bus_dma_tag_destroy(isp->isp_cdmat);
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
free(isp->isp_xflist, M_DEVBUF);
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
#ifdef ISP_TARGET_MODE
free(isp->isp_tgtlist, M_DEVBUF);
#endif
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
free(pcs->dmaps, M_DEVBUF);
Make sure that if are in fact using 'full SMP', make the interrupt flags include INTR_MPSAFE. Put the flags in a common place so that both isp_sbus && isp_pci DTRT. In isp_mbxdma setup, drop any locks prior to calling things like bus_dmatag_create. This gets rid of these obnoxious WITNESS messages about 'sleeping with locks held' blah blah blah blah blah.
2002-07-25 16:02:09 +00:00
ISP_LOCK(isp);
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
return (1);
}
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
for (i = 0; i < isp->isp_maxcmds; i++) {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
error = bus_dmamap_create(pcs->dmat, 0, &pcs->dmaps[i]);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
if (error) {
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
isp_prt(isp, ISP_LOGERR,
"error %d creating per-cmd DMA maps", error);
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
while (--i >= 0) {
bus_dmamap_destroy(pcs->dmat, pcs->dmaps[i]);
}
goto bad;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
}
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
im.isp = isp;
im.error = 0;
bus_dmamap_load(isp->isp_cdmat, isp->isp_cdmap, base, len, imc, &im, 0);
if (im.error) {
isp_prt(isp, ISP_LOGERR,
"error %d loading dma map for control areas", im.error);
goto bad;
}
isp->isp_rquest = base;
base += ISP_QUEUE_SIZE(RQUEST_QUEUE_LEN(isp));
isp->isp_result = base;
Remove pre-CAM code. Add in getenv_int calls for variables isp_mem_map, isp_io_map, isp_no_fwload, isp_fwload, isp_no_nvram, isp_fcduplex which are all bitmaps of isp instances that should or shouldn't map memory space, I/O space, not load f/w, load f/w, ignore nvram, not ignore nvarm, set full duplex mode. Also have an isp_seed value that we can use to generate a pseudo seed for a synthetic WWN. Other minor cosmetic cleanup. Add in support for the Qlogic ISP 2200. Very important change where we actually check now to see whether we were successful in mapping request and response queues (and fibre channel scratch space).
1999-07-02 23:18:03 +00:00
if (IS_FC(isp)) {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
base += ISP_QUEUE_SIZE(RESULT_QUEUE_LEN(isp));
FCPARAM(isp)->isp_scratch = base;
enable 1080 LVD support
1999-03-25 22:53:56 +00:00
}
Make sure that if are in fact using 'full SMP', make the interrupt flags include INTR_MPSAFE. Put the flags in a common place so that both isp_sbus && isp_pci DTRT. In isp_mbxdma setup, drop any locks prior to calling things like bus_dmatag_create. This gets rid of these obnoxious WITNESS messages about 'sleeping with locks held' blah blah blah blah blah.
2002-07-25 16:02:09 +00:00
ISP_LOCK(isp);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
return (0);
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bad:
bus_dmamem_free(isp->isp_cdmat, base, isp->isp_cdmap);
bus_dma_tag_destroy(isp->isp_cdmat);
free(isp->isp_xflist, M_DEVBUF);
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
#ifdef ISP_TARGET_MODE
free(isp->isp_tgtlist, M_DEVBUF);
#endif
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
free(pcs->dmaps, M_DEVBUF);
Make sure that if are in fact using 'full SMP', make the interrupt flags include INTR_MPSAFE. Put the flags in a common place so that both isp_sbus && isp_pci DTRT. In isp_mbxdma setup, drop any locks prior to calling things like bus_dmatag_create. This gets rid of these obnoxious WITNESS messages about 'sleeping with locks held' blah blah blah blah blah.
2002-07-25 16:02:09 +00:00
ISP_LOCK(isp);
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
isp->isp_rquest = NULL;
return (1);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
typedef struct {
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t *isp;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
void *cmd_token;
void *rq;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t *nxtip;
uint16_t optr;
int error;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
} mush_t;
A major amount of cleaning up: + Change some messages about CCB memory allocation + Turn a failure to DMA map all of a transaction due to lack of ISP queue entries into a requeue operation (instead of the case where it had been treated the same as a DMA too big operation). + put back splsoftvm around bus_dmamap_load calls. + cleanup (and fix a glaring bug) in the and of the dma setup routine. Also, the dma setup routines either return CMD_QUEUED (for success) or CMD_COMPLETE (for failure) or CMD_EAGAIN (for requeuing for resource shortage reasons).
1998-09-17 21:11:21 +00:00
#define MUSHERR_NOQENTRIES -2
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
#ifdef ISP_TARGET_MODE
/*
* We need to handle DMA for target mode differently from initiator mode.
*
* DMA mapping and construction and submission of CTIO Request Entries
* and rendevous for completion are very tightly coupled because we start
* out by knowing (per platform) how much data we have to move, but we
* don't know, up front, how many DMA mapping segments will have to be used
* cover that data, so we don't know how many CTIO Request Entries we
* will end up using. Further, for performance reasons we may want to
* (on the last CTIO for Fibre Channel), send status too (if all went well).
*
* The standard vector still goes through isp_pci_dmasetup, but the callback
* for the DMA mapping routines comes here instead with the whole transfer
* mapped and a pointer to a partially filled in already allocated request
* queue entry. We finish the job.
*/
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
static void tdma_mk(void *, bus_dma_segment_t *, int, int);
static void tdma_mkfc(void *, bus_dma_segment_t *, int, int);
#define STATUS_WITH_DATA 1
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
static void
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
tdma_mk(void *arg, bus_dma_segment_t *dm_segs, int nseg, int error)
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
{
mush_t *mp;
struct ccb_scsiio *csio;
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t *isp;
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
struct isp_pcisoftc *pcs;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
bus_dmamap_t *dp;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ct_entry_t *cto, *qe;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t scsi_status;
uint16_t curi, nxti, handle;
uint32_t sflags;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
int32_t resid;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
int nth_ctio, nctios, send_status;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
mp = (mush_t *) arg;
if (error) {
mp->error = error;
return;
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp = mp->isp;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
csio = mp->cmd_token;
cto = mp->rq;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
curi = isp->isp_reqidx;
qe = (ct_entry_t *) ISP_QUEUE_ENTRY(isp->isp_rquest, curi);
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
cto->ct_xfrlen = 0;
cto->ct_seg_count = 0;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
cto->ct_header.rqs_entry_count = 1;
MEMZERO(cto->ct_dataseg, sizeof(cto->ct_dataseg));
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
if (nseg == 0) {
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
cto->ct_header.rqs_seqno = 1;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_prt(isp, ISP_LOGTDEBUG1,
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
"CTIO[%x] lun%d iid%d tag %x flgs %x sts %x ssts %x res %d",
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
cto->ct_fwhandle, csio->ccb_h.target_lun, cto->ct_iid,
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
cto->ct_tag_val, cto->ct_flags, cto->ct_status,
cto->ct_scsi_status, cto->ct_resid);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ISP_TDQE(isp, "tdma_mk[no data]", curi, cto);
isp_put_ctio(isp, cto, qe);
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
return;
}
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
nctios = nseg / ISP_RQDSEG;
if (nseg % ISP_RQDSEG) {
nctios++;
}
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
/*
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
* Save syshandle, and potentially any SCSI status, which we'll
* reinsert on the last CTIO we're going to send.
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
handle = cto->ct_syshandle;
cto->ct_syshandle = 0;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
cto->ct_header.rqs_seqno = 0;
send_status = (cto->ct_flags & CT_SENDSTATUS) != 0;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
if (send_status) {
sflags = cto->ct_flags & (CT_SENDSTATUS | CT_CCINCR);
cto->ct_flags &= ~(CT_SENDSTATUS | CT_CCINCR);
/*
* Preserve residual.
*/
resid = cto->ct_resid;
/*
* Save actual SCSI status.
*/
scsi_status = cto->ct_scsi_status;
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
#ifndef STATUS_WITH_DATA
sflags |= CT_NO_DATA;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
/*
* We can't do a status at the same time as a data CTIO, so
* we need to synthesize an extra CTIO at this level.
*/
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
nctios++;
For parallel SCSI, let us now do status with the final CTIO. For the 1080, I was hanging after sending a xfer CTIO and a status CTIO for a non-discon INQUIRY- the xfer CTIO was returned as completed OK, but the status CTIO was dropped on the floor. All the fields looked good. I don't know why it got dropped. But allowing status to go back with data xfer seemed to work. I also noticed that with a non-disconnecting command that the firmware handle in the ATIO is zero- this leads me to believe that the f/w really can only handle one CTIO at a time in the discon case, and it had no idea what to do with the second (status) CTIO.
2001-03-21 00:49:37 +00:00
#endif
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
} else {
sflags = scsi_status = resid = 0;
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
}
If we're on an interrupt stack, mark things so that we don't try and cv_wait for mailbox commands to complete if we start them from here. Fix residuals for target mode such that we only check the residual and set it in the CTIO if this is the last CTIO (when we're sending status). MFC after: 4 weeks
2001-09-04 21:45:57 +00:00
cto->ct_resid = 0;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
cto->ct_scsi_status = 0;
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
pcs = (struct isp_pcisoftc *)isp;
dp = &pcs->dmaps[isp_handle_index(handle)];
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_dmamap_sync(pcs->dmat, *dp, BUS_DMASYNC_PREREAD);
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
} else {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_dmamap_sync(pcs->dmat, *dp, BUS_DMASYNC_PREWRITE);
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
nxti = *mp->nxtip;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
for (nth_ctio = 0; nth_ctio < nctios; nth_ctio++) {
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
int seglim;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
seglim = nseg;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
if (seglim) {
int seg;
if (seglim > ISP_RQDSEG)
seglim = ISP_RQDSEG;
for (seg = 0; seg < seglim; seg++, nseg--) {
/*
* Unlike normal initiator commands, we don't
* do any swizzling here.
*/
cto->ct_dataseg[seg].ds_count = dm_segs->ds_len;
cto->ct_dataseg[seg].ds_base = dm_segs->ds_addr;
cto->ct_xfrlen += dm_segs->ds_len;
dm_segs++;
}
cto->ct_seg_count = seg;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
} else {
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
/*
* This case should only happen when we're sending an
* extra CTIO with final status.
*/
if (send_status == 0) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_prt(isp, ISP_LOGWARN,
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
"tdma_mk ran out of segments");
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
mp->error = EINVAL;
return;
}
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
}
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
/*
* At this point, the fields ct_lun, ct_iid, ct_tagval,
* ct_tagtype, and ct_timeout have been carried over
* unchanged from what our caller had set.
*
* The dataseg fields and the seg_count fields we just got
* through setting. The data direction we've preserved all
* along and only clear it if we're now sending status.
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (nth_ctio == nctios - 1) {
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
/*
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
* We're the last in a sequence of CTIOs, so mark
* this CTIO and save the handle to the CCB such that
* when this CTIO completes we can free dma resources
* and do whatever else we need to do to finish the
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
* rest of the command. We *don't* give this to the
* firmware to work on- the caller will do that.
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
cto->ct_syshandle = handle;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
cto->ct_header.rqs_seqno = 1;
if (send_status) {
cto->ct_scsi_status = scsi_status;
For parallel SCSI, let us now do status with the final CTIO. For the 1080, I was hanging after sending a xfer CTIO and a status CTIO for a non-discon INQUIRY- the xfer CTIO was returned as completed OK, but the status CTIO was dropped on the floor. All the fields looked good. I don't know why it got dropped. But allowing status to go back with data xfer seemed to work. I also noticed that with a non-disconnecting command that the firmware handle in the ATIO is zero- this leads me to believe that the f/w really can only handle one CTIO at a time in the discon case, and it had no idea what to do with the second (status) CTIO.
2001-03-21 00:49:37 +00:00
cto->ct_flags |= sflags;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
cto->ct_resid = resid;
}
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
if (send_status) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_prt(isp, ISP_LOGTDEBUG1,
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
"CTIO[%x] lun%d iid %d tag %x ct_flags %x "
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
"scsi status %x resid %d",
cto->ct_fwhandle, csio->ccb_h.target_lun,
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
cto->ct_iid, cto->ct_tag_val, cto->ct_flags,
Add in an isp_tdebug environment variable. Clean up some debugging printouts for clarity.
2000-01-04 03:41:40 +00:00
cto->ct_scsi_status, cto->ct_resid);
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
} else {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_prt(isp, ISP_LOGTDEBUG1,
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
"CTIO[%x] lun%d iid%d tag %x ct_flags 0x%x",
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
cto->ct_fwhandle, csio->ccb_h.target_lun,
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
cto->ct_iid, cto->ct_tag_val,
cto->ct_flags);
Add in an isp_tdebug environment variable. Clean up some debugging printouts for clarity.
2000-01-04 03:41:40 +00:00
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_put_ctio(isp, cto, qe);
ISP_TDQE(isp, "last tdma_mk", curi, cto);
if (nctios > 1) {
MEMORYBARRIER(isp, SYNC_REQUEST,
curi, QENTRY_LEN);
}
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
} else {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ct_entry_t *oqe = qe;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
/*
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
* Make sure syshandle fields are clean
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
*/
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
cto->ct_syshandle = 0;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
cto->ct_header.rqs_seqno = 0;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_prt(isp, ISP_LOGTDEBUG1,
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
"CTIO[%x] lun%d for ID%d ct_flags 0x%x",
cto->ct_fwhandle, csio->ccb_h.target_lun,
cto->ct_iid, cto->ct_flags);
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
/*
* Get a new CTIO
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
qe = (ct_entry_t *)
ISP_QUEUE_ENTRY(isp->isp_rquest, nxti);
nxti = ISP_NXT_QENTRY(nxti, RQUEST_QUEUE_LEN(isp));
if (nxti == mp->optr) {
isp_prt(isp, ISP_LOGTDEBUG0,
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
"Queue Overflow in tdma_mk");
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
mp->error = MUSHERR_NOQENTRIES;
return;
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
/*
* Now that we're done with the old CTIO,
* flush it out to the request queue.
*/
ISP_TDQE(isp, "dma_tgt_fc", curi, cto);
isp_put_ctio(isp, cto, oqe);
if (nth_ctio != 0) {
MEMORYBARRIER(isp, SYNC_REQUEST, curi,
QENTRY_LEN);
}
curi = ISP_NXT_QENTRY(curi, RQUEST_QUEUE_LEN(isp));
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
/*
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
* Reset some fields in the CTIO so we can reuse
* for the next one we'll flush to the request
* queue.
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
*/
cto->ct_header.rqs_entry_type = RQSTYPE_CTIO;
cto->ct_header.rqs_entry_count = 1;
cto->ct_header.rqs_flags = 0;
cto->ct_status = 0;
cto->ct_scsi_status = 0;
cto->ct_xfrlen = 0;
cto->ct_resid = 0;
cto->ct_seg_count = 0;
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
MEMZERO(cto->ct_dataseg, sizeof(cto->ct_dataseg));
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
}
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
*mp->nxtip = nxti;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
}
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
/*
* We don't have to do multiple CTIOs here. Instead, we can just do
* continuation segments as needed. This greatly simplifies the code
* improves performance.
*/
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
static void
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
tdma_mkfc(void *arg, bus_dma_segment_t *dm_segs, int nseg, int error)
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
{
mush_t *mp;
struct ccb_scsiio *csio;
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t *isp;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ct2_entry_t *cto, *qe;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t curi, nxti;
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
ispds_t *ds;
ispds64_t *ds64;
int segcnt, seglim;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
mp = (mush_t *) arg;
if (error) {
mp->error = error;
return;
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp = mp->isp;
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
csio = mp->cmd_token;
cto = mp->rq;
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
curi = isp->isp_reqidx;
qe = (ct2_entry_t *) ISP_QUEUE_ENTRY(isp->isp_rquest, curi);
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
if (nseg == 0) {
if ((cto->ct_flags & CT2_FLAG_MMASK) != CT2_FLAG_MODE1) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_prt(isp, ISP_LOGWARN,
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
"dma2_tgt_fc, a status CTIO2 without MODE1 "
"set (0x%x)", cto->ct_flags);
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
mp->error = EINVAL;
return;
}
/*
* We preserve ct_lun, ct_iid, ct_rxid. We set the data
* flags to NO DATA and clear relative offset flags.
* We preserve the ct_resid and the response area.
*/
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
cto->ct_header.rqs_seqno = 1;
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
cto->ct_seg_count = 0;
cto->ct_reloff = 0;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_prt(isp, ISP_LOGTDEBUG1,
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
"CTIO2[%x] lun %d->iid%d flgs 0x%x sts 0x%x ssts "
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
"0x%x res %d", cto->ct_rxid, csio->ccb_h.target_lun,
cto->ct_iid, cto->ct_flags, cto->ct_status,
cto->rsp.m1.ct_scsi_status, cto->ct_resid);
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
if (IS_2KLOGIN(isp)) {
isp_put_ctio2e(isp,
(ct2e_entry_t *)cto, (ct2e_entry_t *)qe);
} else {
isp_put_ctio2(isp, cto, qe);
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ISP_TDQE(isp, "dma2_tgt_fc[no data]", curi, qe);
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
return;
}
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
if ((cto->ct_flags & CT2_FLAG_MMASK) != CT2_FLAG_MODE0) {
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
isp_prt(isp, ISP_LOGERR,
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
"dma2_tgt_fc, a data CTIO2 without MODE0 set "
"(0x%x)", cto->ct_flags);
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
mp->error = EINVAL;
return;
}
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
nxti = *mp->nxtip;
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
/*
* Check to see if we need to DAC addressing or not.
*
* Any address that's over the 4GB boundary causes this
* to happen.
*/
segcnt = nseg;
if (sizeof (bus_addr_t) > 4) {
for (segcnt = 0; segcnt < nseg; segcnt++) {
uint64_t addr = dm_segs[segcnt].ds_addr;
if (addr >= 0x100000000LL) {
break;
}
}
}
if (segcnt != nseg) {
cto->ct_header.rqs_entry_type = RQSTYPE_CTIO3;
seglim = ISP_RQDSEG_T3;
ds64 = &cto->rsp.m0.ct_dataseg64[0];
ds = NULL;
} else {
seglim = ISP_RQDSEG_T2;
ds64 = NULL;
ds = &cto->rsp.m0.ct_dataseg[0];
}
cto->ct_seg_count = 0;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
/*
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
* Set up the CTIO2 data segments.
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
*/
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
for (segcnt = 0; cto->ct_seg_count < seglim && segcnt < nseg;
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
cto->ct_seg_count++, segcnt++) {
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
if (ds64) {
ds64->ds_basehi =
((uint64_t) (dm_segs[segcnt].ds_addr) >> 32);
ds64->ds_base = dm_segs[segcnt].ds_addr;
ds64->ds_count = dm_segs[segcnt].ds_len;
ds64++;
} else {
ds->ds_base = dm_segs[segcnt].ds_addr;
ds->ds_count = dm_segs[segcnt].ds_len;
ds++;
}
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
cto->rsp.m0.ct_xfrlen += dm_segs[segcnt].ds_len;
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#if __FreeBSD_version < 500000
isp_prt(isp, ISP_LOGTDEBUG1,
"isp_send_ctio2: ent0[%d]0x%llx:%llu",
cto->ct_seg_count, (uint64_t)dm_segs[segcnt].ds_addr,
(uint64_t)dm_segs[segcnt].ds_len);
#else
Make compiles (LINT and/or ISP_TARGET_MODE options) happier by making sure printf type format args and actual args match. Reviewed by: Sam Leffler <sam@errno.com>
2003-01-02 04:21:52 +00:00
isp_prt(isp, ISP_LOGTDEBUG1,
Use uintmax_t and %j to print bus dma segment members rather than casting to long long and using %ll.
2005-12-15 22:12:27 +00:00
"isp_send_ctio2: ent0[%d]0x%jx:%ju",
cto->ct_seg_count, (uintmax_t)dm_segs[segcnt].ds_addr,
(uintmax_t)dm_segs[segcnt].ds_len);
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#endif
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
}
while (segcnt < nseg) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t curip;
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
int seg;
ispcontreq_t local, *crq = &local, *qep;
qep = (ispcontreq_t *) ISP_QUEUE_ENTRY(isp->isp_rquest, nxti);
curip = nxti;
nxti = ISP_NXT_QENTRY(curip, RQUEST_QUEUE_LEN(isp));
if (nxti == mp->optr) {
ISP_UNLOCK(isp);
isp_prt(isp, ISP_LOGTDEBUG0,
"tdma_mkfc: request queue overflow");
mp->error = MUSHERR_NOQENTRIES;
return;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
}
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
cto->ct_header.rqs_entry_count++;
MEMZERO((void *)crq, sizeof (*crq));
crq->req_header.rqs_entry_count = 1;
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
if (cto->ct_header.rqs_entry_type == RQSTYPE_CTIO3) {
seglim = ISP_CDSEG64;
ds = NULL;
ds64 = &((ispcontreq64_t *)crq)->req_dataseg[0];
crq->req_header.rqs_entry_type = RQSTYPE_A64_CONT;
} else {
seglim = ISP_CDSEG;
ds = &crq->req_dataseg[0];
ds64 = NULL;
crq->req_header.rqs_entry_type = RQSTYPE_DATASEG;
}
for (seg = 0; segcnt < nseg && seg < seglim;
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
segcnt++, seg++) {
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
if (ds64) {
ds64->ds_basehi =
((uint64_t) (dm_segs[segcnt].ds_addr) >> 32);
ds64->ds_base = dm_segs[segcnt].ds_addr;
ds64->ds_count = dm_segs[segcnt].ds_len;
ds64++;
} else {
ds->ds_base = dm_segs[segcnt].ds_addr;
ds->ds_count = dm_segs[segcnt].ds_len;
ds++;
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#if __FreeBSD_version < 500000
isp_prt(isp, ISP_LOGTDEBUG1,
"isp_send_ctio2: ent%d[%d]%llx:%llu",
cto->ct_header.rqs_entry_count-1, seg,
(uint64_t)dm_segs[segcnt].ds_addr,
(uint64_t)dm_segs[segcnt].ds_len);
#else
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_prt(isp, ISP_LOGTDEBUG1,
Use uintmax_t and %j to print bus dma segment members rather than casting to long long and using %ll.
2005-12-15 22:12:27 +00:00
"isp_send_ctio2: ent%d[%d]%jx:%ju",
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
cto->ct_header.rqs_entry_count-1, seg,
Use uintmax_t and %j to print bus dma segment members rather than casting to long long and using %ll.
2005-12-15 22:12:27 +00:00
(uintmax_t)dm_segs[segcnt].ds_addr,
(uintmax_t)dm_segs[segcnt].ds_len);
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
#endif
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
cto->rsp.m0.ct_xfrlen += dm_segs[segcnt].ds_len;
cto->ct_seg_count++;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
}
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
MEMORYBARRIER(isp, SYNC_REQUEST, curip, QENTRY_LEN);
isp_put_cont_req(isp, crq, qep);
ISP_TDQE(isp, "cont entry", curi, qep);
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
}
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
/*
* No do final twiddling for the CTIO itself.
*/
cto->ct_header.rqs_seqno = 1;
isp_prt(isp, ISP_LOGTDEBUG1,
"CTIO2[%x] lun %d->iid%d flgs 0x%x sts 0x%x ssts 0x%x resid %d",
cto->ct_rxid, csio->ccb_h.target_lun, (int) cto->ct_iid,
cto->ct_flags, cto->ct_status, cto->rsp.m1.ct_scsi_status,
cto->ct_resid);
Remove bzero/bcopy vestiges Be cognizant as to whether we're running 2KLogin f/w in target mode and do the appropriate loopid load based upon that. Do a first cut (seems to work, at least for amd64) at 64 bit target mode for fibre channel cards. We could probably also do it for SPI cards, but that's not supported right now.
2006-05-22 06:51:48 +00:00
if (IS_2KLOGIN(isp))
isp_put_ctio2e(isp, (ct2e_entry_t *)cto, (ct2e_entry_t *)qe);
else
isp_put_ctio2(isp, cto, qe);
Set all 23XX cards as 'touched' (we have trouble, unpredictably, about running ABOUT FIRMWARE with some that were started by BIOS downloads). Redo CTIO2 dma mapping- use continuation segments instead of multiple CTIO2s. Thanks to Veritas for sponsoring this work (in a different context). MFC after: 1 week
2002-06-16 04:58:00 +00:00
ISP_TDQE(isp, "last dma2_tgt_fc", curi, qe);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
*mp->nxtip = nxti;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
}
#endif
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
static void dma2_a64(void *, bus_dma_segment_t *, int, int);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
static void dma2(void *, bus_dma_segment_t *, int, int);
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
static void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
dma2_a64(void *arg, bus_dma_segment_t *dm_segs, int nseg, int error)
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
{
mush_t *mp;
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t *isp;
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
struct ccb_scsiio *csio;
struct isp_pcisoftc *pcs;
bus_dmamap_t *dp;
bus_dma_segment_t *eseg;
ispreq64_t *rq;
int seglim, datalen;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t nxti;
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
mp = (mush_t *) arg;
if (error) {
mp->error = error;
return;
}
if (nseg < 1) {
isp_prt(mp->isp, ISP_LOGERR, "bad segment count (%d)", nseg);
mp->error = EFAULT;
return;
}
csio = mp->cmd_token;
isp = mp->isp;
rq = mp->rq;
pcs = (struct isp_pcisoftc *)mp->isp;
dp = &pcs->dmaps[isp_handle_index(rq->req_handle)];
nxti = *mp->nxtip;
if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
bus_dmamap_sync(pcs->dmat, *dp, BUS_DMASYNC_PREREAD);
} else {
bus_dmamap_sync(pcs->dmat, *dp, BUS_DMASYNC_PREWRITE);
}
datalen = XS_XFRLEN(csio);
/*
* We're passed an initial partially filled in entry that
* has most fields filled in except for data transfer
* related values.
*
* Our job is to fill in the initial request queue entry and
* then to start allocating and filling in continuation entries
* until we've covered the entire transfer.
*/
if (IS_FC(isp)) {
PAE support changes that included at least some minimal actual testing with a kernel that booted.
2004-09-23 05:25:22 +00:00
rq->req_header.rqs_entry_type = RQSTYPE_T3RQS;
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
seglim = ISP_RQDSEG_T3;
((ispreqt3_t *)rq)->req_totalcnt = datalen;
if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
((ispreqt3_t *)rq)->req_flags |= REQFLAG_DATA_IN;
} else {
((ispreqt3_t *)rq)->req_flags |= REQFLAG_DATA_OUT;
}
} else {
PAE support changes that included at least some minimal actual testing with a kernel that booted.
2004-09-23 05:25:22 +00:00
rq->req_header.rqs_entry_type = RQSTYPE_A64;
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
if (csio->cdb_len > 12) {
seglim = 0;
} else {
seglim = ISP_RQDSEG_A64;
}
if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
rq->req_flags |= REQFLAG_DATA_IN;
} else {
rq->req_flags |= REQFLAG_DATA_OUT;
}
}
eseg = dm_segs + nseg;
while (datalen != 0 && rq->req_seg_count < seglim && dm_segs != eseg) {
if (IS_FC(isp)) {
ispreqt3_t *rq3 = (ispreqt3_t *)rq;
rq3->req_dataseg[rq3->req_seg_count].ds_base =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
DMA_LO32(dm_segs->ds_addr);
PAE support changes that included at least some minimal actual testing with a kernel that booted.
2004-09-23 05:25:22 +00:00
rq3->req_dataseg[rq3->req_seg_count].ds_basehi =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
DMA_HI32(dm_segs->ds_addr);
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
rq3->req_dataseg[rq3->req_seg_count].ds_count =
dm_segs->ds_len;
} else {
rq->req_dataseg[rq->req_seg_count].ds_base =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
DMA_LO32(dm_segs->ds_addr);
PAE support changes that included at least some minimal actual testing with a kernel that booted.
2004-09-23 05:25:22 +00:00
rq->req_dataseg[rq->req_seg_count].ds_basehi =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
DMA_HI32(dm_segs->ds_addr);
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
rq->req_dataseg[rq->req_seg_count].ds_count =
PAE support changes that included at least some minimal actual testing with a kernel that booted.
2004-09-23 05:25:22 +00:00
dm_segs->ds_len;
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
}
datalen -= dm_segs->ds_len;
rq->req_seg_count++;
dm_segs++;
}
while (datalen > 0 && dm_segs != eseg) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t onxti;
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
ispcontreq64_t local, *crq = &local, *cqe;
cqe = (ispcontreq64_t *) ISP_QUEUE_ENTRY(isp->isp_rquest, nxti);
onxti = nxti;
nxti = ISP_NXT_QENTRY(onxti, RQUEST_QUEUE_LEN(isp));
if (nxti == mp->optr) {
isp_prt(isp, ISP_LOGDEBUG0, "Request Queue Overflow++");
mp->error = MUSHERR_NOQENTRIES;
return;
}
rq->req_header.rqs_entry_count++;
MEMZERO((void *)crq, sizeof (*crq));
crq->req_header.rqs_entry_count = 1;
crq->req_header.rqs_entry_type = RQSTYPE_A64_CONT;
seglim = 0;
while (datalen > 0 && seglim < ISP_CDSEG64 && dm_segs != eseg) {
crq->req_dataseg[seglim].ds_base =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
DMA_LO32(dm_segs->ds_addr);
PAE support changes that included at least some minimal actual testing with a kernel that booted.
2004-09-23 05:25:22 +00:00
crq->req_dataseg[seglim].ds_basehi =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
DMA_HI32(dm_segs->ds_addr);
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
crq->req_dataseg[seglim].ds_count =
dm_segs->ds_len;
rq->req_seg_count++;
dm_segs++;
seglim++;
datalen -= dm_segs->ds_len;
}
isp_put_cont64_req(isp, crq, cqe);
MEMORYBARRIER(isp, SYNC_REQUEST, onxti, QENTRY_LEN);
}
*mp->nxtip = nxti;
}
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
static void
clarify headers;ansify
1998-12-28 19:24:23 +00:00
dma2(void *arg, bus_dma_segment_t *dm_segs, int nseg, int error)
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
{
mush_t *mp;
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t *isp;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
struct ccb_scsiio *csio;
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
struct isp_pcisoftc *pcs;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
bus_dmamap_t *dp;
bus_dma_segment_t *eseg;
ispreq_t *rq;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
int seglim, datalen;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t nxti;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
mp = (mush_t *) arg;
if (error) {
mp->error = error;
return;
}
if (nseg < 1) {
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
isp_prt(mp->isp, ISP_LOGERR, "bad segment count (%d)", nseg);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
mp->error = EFAULT;
return;
}
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
csio = mp->cmd_token;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp = mp->isp;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
rq = mp->rq;
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
pcs = (struct isp_pcisoftc *)mp->isp;
dp = &pcs->dmaps[isp_handle_index(rq->req_handle)];
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
nxti = *mp->nxtip;
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_dmamap_sync(pcs->dmat, *dp, BUS_DMASYNC_PREREAD);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
} else {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_dmamap_sync(pcs->dmat, *dp, BUS_DMASYNC_PREWRITE);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
datalen = XS_XFRLEN(csio);
/*
* We're passed an initial partially filled in entry that
* has most fields filled in except for data transfer
* related values.
*
* Our job is to fill in the initial request queue entry and
* then to start allocating and filling in continuation entries
* until we've covered the entire transfer.
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (IS_FC(isp)) {
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
seglim = ISP_RQDSEG_T2;
((ispreqt2_t *)rq)->req_totalcnt = datalen;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
((ispreqt2_t *)rq)->req_flags |= REQFLAG_DATA_IN;
} else {
((ispreqt2_t *)rq)->req_flags |= REQFLAG_DATA_OUT;
}
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
} else {
If the CDB length is greater than 12 for parallel SCSI, ispscsicmd has made the initial queue entry a EXTENDED CMD queue entry, so we have to go straight to continuation segments for any data segments. approved: jkh
2000-02-15 00:50:01 +00:00
if (csio->cdb_len > 12) {
seglim = 0;
} else {
seglim = ISP_RQDSEG;
}
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
rq->req_flags |= REQFLAG_DATA_IN;
} else {
rq->req_flags |= REQFLAG_DATA_OUT;
}
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
eseg = dm_segs + nseg;
while (datalen != 0 && rq->req_seg_count < seglim && dm_segs != eseg) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (IS_FC(isp)) {
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
ispreqt2_t *rq2 = (ispreqt2_t *)rq;
rq2->req_dataseg[rq2->req_seg_count].ds_base =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
DMA_LO32(dm_segs->ds_addr);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
rq2->req_dataseg[rq2->req_seg_count].ds_count =
dm_segs->ds_len;
} else {
rq->req_dataseg[rq->req_seg_count].ds_base =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
DMA_LO32(dm_segs->ds_addr);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
rq->req_dataseg[rq->req_seg_count].ds_count =
dm_segs->ds_len;
}
datalen -= dm_segs->ds_len;
rq->req_seg_count++;
dm_segs++;
}
while (datalen > 0 && dm_segs != eseg) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t onxti;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ispcontreq_t local, *crq = &local, *cqe;
cqe = (ispcontreq_t *) ISP_QUEUE_ENTRY(isp->isp_rquest, nxti);
onxti = nxti;
nxti = ISP_NXT_QENTRY(onxti, RQUEST_QUEUE_LEN(isp));
if (nxti == mp->optr) {
isp_prt(isp, ISP_LOGDEBUG0, "Request Queue Overflow++");
A major amount of cleaning up: + Change some messages about CCB memory allocation + Turn a failure to DMA map all of a transaction due to lack of ISP queue entries into a requeue operation (instead of the case where it had been treated the same as a DMA too big operation). + put back splsoftvm around bus_dmamap_load calls. + cleanup (and fix a glaring bug) in the and of the dma setup routine. Also, the dma setup routines either return CMD_QUEUED (for success) or CMD_COMPLETE (for failure) or CMD_EAGAIN (for requeuing for resource shortage reasons).
1998-09-17 21:11:21 +00:00
mp->error = MUSHERR_NOQENTRIES;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
return;
}
rq->req_header.rqs_entry_count++;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
MEMZERO((void *)crq, sizeof (*crq));
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
crq->req_header.rqs_entry_count = 1;
crq->req_header.rqs_entry_type = RQSTYPE_DATASEG;
seglim = 0;
while (datalen > 0 && seglim < ISP_CDSEG && dm_segs != eseg) {
crq->req_dataseg[seglim].ds_base =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
DMA_LO32(dm_segs->ds_addr);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
crq->req_dataseg[seglim].ds_count =
dm_segs->ds_len;
rq->req_seg_count++;
dm_segs++;
seglim++;
datalen -= dm_segs->ds_len;
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_put_cont_req(isp, crq, cqe);
MEMORYBARRIER(isp, SYNC_REQUEST, onxti, QENTRY_LEN);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
*mp->nxtip = nxti;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
/*
* We enter with ISP_LOCK held
*/
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_dmasetup(ispsoftc_t *isp, struct ccb_scsiio *csio, ispreq_t *rq,
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t *nxtip, uint16_t optr)
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
{
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *)isp;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ispreq_t *qep;
Set some correct return values. Prefer I/O map all the time unless configured otherwise.
1999-08-16 19:52:29 +00:00
bus_dmamap_t *dp = NULL;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
mush_t mush, *mp;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
void (*eptr)(void *, bus_dma_segment_t *, int, int);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
qep = (ispreq_t *) ISP_QUEUE_ENTRY(isp->isp_rquest, isp->isp_reqidx);
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
#ifdef ISP_TARGET_MODE
if (csio->ccb_h.func_code == XPT_CONT_TARGET_IO) {
if (IS_FC(isp)) {
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
eptr = tdma_mkfc;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
} else {
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
eptr = tdma_mk;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
}
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE ||
(csio->dxfer_len == 0)) {
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
mp = &mush;
mp->isp = isp;
mp->cmd_token = csio;
Complete some Ansification. Check to make sure, in tdma_mk, that we won't overflow the request queue. The reason we want to do this is that we now push out completed CTIOs as we complete them- this gets the QLogic working on them quicker. So we need to know whether we can put the entire burrito out before we start. We now support conjoint status with data for the last CTIO for both Fibre Channel and SCSI. Leave the old code in place in case we need to go back (minor 3 line ifdef). Ultra-ultra important- *don't* set rq->req_seg_count for non-data target mode requests in isp_pci_dmasetup. D'oh- this is actually the tag value area for a CTIO. What *was* I thinking? Boy howdy does both aic7xxx and sym get awfully unhappy when on reconnect you give them a constant '1' for a tag value.
2001-04-04 21:53:59 +00:00
mp->rq = rq; /* really a ct_entry_t or ct2_entry_t */
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
mp->nxtip = nxtip;
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
mp->optr = optr;
mp->error = 0;
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ISPLOCK_2_CAMLOCK(isp);
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
(*eptr)(mp, NULL, 0, 0);
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
CAMLOCK_2_ISPLOCK(isp);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
goto mbxsync;
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
}
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
} else
#endif
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
if (sizeof (bus_addr_t) > 4) {
eptr = dma2_a64;
} else {
eptr = dma2;
}
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
Keep interrupts blocked for all of isp_pci_attach. Redo DMA routines for target mode for cleanliness and accuracy.
2000-07-18 06:40:22 +00:00
if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE ||
(csio->dxfer_len == 0)) {
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
rq->req_seg_count = 1;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
goto mbxsync;
Redo FC target mode dma routine to try and generate an extra CTIO in the not so odd case of Moving Data *AND* Sending Status in last CTIO *AND* status is a CHECK CONDITION *AND* we have Sense Data to send.
2000-01-15 01:54:36 +00:00
}
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
/*
* Do a virtual grapevine step to collect info for
A major amount of cleaning up: + Change some messages about CCB memory allocation + Turn a failure to DMA map all of a transaction due to lack of ISP queue entries into a requeue operation (instead of the case where it had been treated the same as a DMA too big operation). + put back splsoftvm around bus_dmamap_load calls. + cleanup (and fix a glaring bug) in the and of the dma setup routine. Also, the dma setup routines either return CMD_QUEUED (for success) or CMD_COMPLETE (for failure) or CMD_EAGAIN (for requeuing for resource shortage reasons).
1998-09-17 21:11:21 +00:00
* the callback dma allocation that we have to use...
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
*/
mp = &mush;
mp->isp = isp;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
mp->cmd_token = csio;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
mp->rq = rq;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
mp->nxtip = nxtip;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
mp->optr = optr;
mp->error = 0;
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ISPLOCK_2_CAMLOCK(isp);
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
if ((csio->ccb_h.flags & CAM_SCATTER_VALID) == 0) {
if ((csio->ccb_h.flags & CAM_DATA_PHYS) == 0) {
A major amount of cleaning up: + Change some messages about CCB memory allocation + Turn a failure to DMA map all of a transaction due to lack of ISP queue entries into a requeue operation (instead of the case where it had been treated the same as a DMA too big operation). + put back splsoftvm around bus_dmamap_load calls. + cleanup (and fix a glaring bug) in the and of the dma setup routine. Also, the dma setup routines either return CMD_QUEUED (for success) or CMD_COMPLETE (for failure) or CMD_EAGAIN (for requeuing for resource shortage reasons).
1998-09-17 21:11:21 +00:00
int error, s;
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
dp = &pcs->dmaps[isp_handle_index(rq->req_handle)];
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
s = splsoftvm();
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
error = bus_dmamap_load(pcs->dmat, *dp,
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
csio->data_ptr, csio->dxfer_len, eptr, mp, 0);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
if (error == EINPROGRESS) {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_dmamap_unload(pcs->dmat, *dp);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
mp->error = EINVAL;
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
isp_prt(isp, ISP_LOGERR,
"deferred dma allocation not supported");
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
} else if (error && mp->error == 0) {
Set some correct return values. Prefer I/O map all the time unless configured otherwise.
1999-08-16 19:52:29 +00:00
#ifdef DIAGNOSTIC
Eliminate the use of the getenv_int stuff we'd been using (with a bitmap for selecting unit). Instead, use the resource hints mechanism. One unfortunate situation here is that there is no resource_quad_value function- which is what I needed for WWN boot time replacement. Worse- you can't store the hint as just plain hint.isp.0.nodewwn="0x50000000aaaa0001" because this gets interpreted as an int- incorrectly because it can't be converted to an int. I can't even get this as a string. To work around this particular case for nodewwn && portwwn setting, this rather grotesque form will be used: hint.isp.0.nodewwn="w50000000aaaa0001" hint.isp.0.portwwn="w50000000aaaa0002" At the same time, if we have no hinted WWN, set the default WWN (which, btw, gets overridden if the card has valid NVRAM, which is usual) to 0x400000007F000009ull (which translates to NAA == IPv4, 127.0.0.9). Eliminate more printf's and replace them either with device_printf or isp_prt calls.
2001-03-01 02:21:36 +00:00
isp_prt(isp, ISP_LOGERR,
"error %d in dma mapping code", error);
Set some correct return values. Prefer I/O map all the time unless configured otherwise.
1999-08-16 19:52:29 +00:00
#endif
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
mp->error = error;
}
A major amount of cleaning up: + Change some messages about CCB memory allocation + Turn a failure to DMA map all of a transaction due to lack of ISP queue entries into a requeue operation (instead of the case where it had been treated the same as a DMA too big operation). + put back splsoftvm around bus_dmamap_load calls. + cleanup (and fix a glaring bug) in the and of the dma setup routine. Also, the dma setup routines either return CMD_QUEUED (for success) or CMD_COMPLETE (for failure) or CMD_EAGAIN (for requeuing for resource shortage reasons).
1998-09-17 21:11:21 +00:00
splx(s);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
} else {
/* Pointer to physical buffer */
Prep for 1080/1240 support. Those fine h/w engineers at Qlogic gave yet another internal register layout model for what is *still* the same architecture. I hope they saved billyuns of gates 'coz otherwise this is *really* annoying.
1999-03-17 05:07:18 +00:00
struct bus_dma_segment seg;
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
seg.ds_addr = (bus_addr_t)(vm_offset_t)csio->data_ptr;
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
seg.ds_len = csio->dxfer_len;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
(*eptr)(mp, &seg, 1, 0);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
} else {
struct bus_dma_segment *segs;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
if ((csio->ccb_h.flags & CAM_DATA_PHYS) != 0) {
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
isp_prt(isp, ISP_LOGERR,
"Physical segment pointers unsupported");
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
mp->error = EINVAL;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
} else if ((csio->ccb_h.flags & CAM_SG_LIST_PHYS) == 0) {
Remove more printfs and use either isp_prt or device_printf. Remember to set ISP_LOGINFO if bootverbose is set.
2000-12-05 07:38:41 +00:00
isp_prt(isp, ISP_LOGERR,
"Virtual segment addresses unsupported");
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
mp->error = EINVAL;
} else {
/* Just use the segments provided */
segs = (struct bus_dma_segment *) csio->data_ptr;
Support target mode operations. This involves having some variant dma mapping callback routines to select from as target mode entries are handled a fair bit differently from normal initiator mode entries.
2000-01-03 23:50:13 +00:00
(*eptr)(mp, segs, csio->sglist_cnt, 0);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
CAMLOCK_2_ISPLOCK(isp);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
if (mp->error) {
A major amount of cleaning up: + Change some messages about CCB memory allocation + Turn a failure to DMA map all of a transaction due to lack of ISP queue entries into a requeue operation (instead of the case where it had been treated the same as a DMA too big operation). + put back splsoftvm around bus_dmamap_load calls. + cleanup (and fix a glaring bug) in the and of the dma setup routine. Also, the dma setup routines either return CMD_QUEUED (for success) or CMD_COMPLETE (for failure) or CMD_EAGAIN (for requeuing for resource shortage reasons).
1998-09-17 21:11:21 +00:00
int retval = CMD_COMPLETE;
if (mp->error == MUSHERR_NOQENTRIES) {
retval = CMD_EAGAIN;
} else if (mp->error == EFBIG) {
Set some correct return values. Prefer I/O map all the time unless configured otherwise.
1999-08-16 19:52:29 +00:00
XS_SETERR(csio, CAM_REQ_TOO_BIG);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
} else if (mp->error == EINVAL) {
Set some correct return values. Prefer I/O map all the time unless configured otherwise.
1999-08-16 19:52:29 +00:00
XS_SETERR(csio, CAM_REQ_INVALID);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
} else {
Set some correct return values. Prefer I/O map all the time unless configured otherwise.
1999-08-16 19:52:29 +00:00
XS_SETERR(csio, CAM_UNREC_HBA_ERROR);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
A major amount of cleaning up: + Change some messages about CCB memory allocation + Turn a failure to DMA map all of a transaction due to lack of ISP queue entries into a requeue operation (instead of the case where it had been treated the same as a DMA too big operation). + put back splsoftvm around bus_dmamap_load calls. + cleanup (and fix a glaring bug) in the and of the dma setup routine. Also, the dma setup routines either return CMD_QUEUED (for success) or CMD_COMPLETE (for failure) or CMD_EAGAIN (for requeuing for resource shortage reasons).
1998-09-17 21:11:21 +00:00
return (retval);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
mbxsync:
switch (rq->req_header.rqs_entry_type) {
case RQSTYPE_REQUEST:
isp_put_request(isp, rq, qep);
break;
case RQSTYPE_CMDONLY:
isp_put_extended_request(isp, (ispextreq_t *)rq,
(ispextreq_t *)qep);
break;
case RQSTYPE_T2RQS:
isp_put_request_t2(isp, (ispreqt2_t *) rq, (ispreqt2_t *) qep);
break;
Do the small amount of tweaking to support PAE for at least initiator mode. I was unable to test this as the PAE kernel crashed with a "cannot copy LDT" before coming up. When this gets a bit more testing, I'll fix the PAE conf file to allow isp devices. PR: 59728
2004-09-07 08:04:09 +00:00
case RQSTYPE_A64:
case RQSTYPE_T3RQS:
isp_put_request_t3(isp, (ispreqt3_t *) rq, (ispreqt3_t *) qep);
break;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
return (CMD_QUEUED);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_dmateardown(ispsoftc_t *isp, XS_T *xs, uint16_t handle)
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
{
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *)isp;
bus_dmamap_t *dp = &pcs->dmaps[isp_handle_index(handle)];
Add in inclusion of machine/md_var.h (so alpha_scsi_bus_register or what have you is prototyped). Removed code versions in md struct- not used any more. Allocate transfer dma maps and xflist stuff in mbxdmasetup based upon isp->isp_maxcmds. Allow for multiple calls to mbxdmasetup (for isp_reset cases).
1999-10-17 19:03:11 +00:00
if ((xs->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_dmamap_sync(pcs->dmat, *dp, BUS_DMASYNC_POSTREAD);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
} else {
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_dmamap_sync(pcs->dmat, *dp, BUS_DMASYNC_POSTWRITE);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
bus_dmamap_unload(pcs->dmat, *dp);
Convert ISP pci front end to CAM/bus space/dma. Convert ncr driver to CAM.
1998-09-15 10:06:23 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_reset1(ispsoftc_t *isp)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
{
/* Make sure the BIOS is disabled */
isp_pci_wr_reg(isp, HCCR, PCI_HCCR_CMD_BIOS);
Change startup locking. Use new isp_handle_index function for indexing off of handles to get dma maps.
2000-07-04 01:01:15 +00:00
/* and enable interrupts */
ENABLE_INTS(isp);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pci_dumpregs(ispsoftc_t *isp, const char *msg)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
{
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
struct isp_pcisoftc *pcs = (struct isp_pcisoftc *)isp;
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
if (msg)
Eliminate the use of the getenv_int stuff we'd been using (with a bitmap for selecting unit). Instead, use the resource hints mechanism. One unfortunate situation here is that there is no resource_quad_value function- which is what I needed for WWN boot time replacement. Worse- you can't store the hint as just plain hint.isp.0.nodewwn="0x50000000aaaa0001" because this gets interpreted as an int- incorrectly because it can't be converted to an int. I can't even get this as a string. To work around this particular case for nodewwn && portwwn setting, this rather grotesque form will be used: hint.isp.0.nodewwn="w50000000aaaa0001" hint.isp.0.portwwn="w50000000aaaa0002" At the same time, if we have no hinted WWN, set the default WWN (which, btw, gets overridden if the card has valid NVRAM, which is usual) to 0x400000007F000009ull (which translates to NAA == IPv4, 127.0.0.9). Eliminate more printf's and replace them either with device_printf or isp_prt calls.
2001-03-01 02:21:36 +00:00
printf("%s: %s\n", device_get_nameunit(isp->isp_dev), msg);
else
printf("%s:\n", device_get_nameunit(isp->isp_dev));
Part of major rewrite for core version 2.0- clarification of mdvec structure, removal of printf/CFGPRINTF in place of isp_prt calls. Parameterization of RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN.
2000-08-01 05:16:49 +00:00
if (IS_SCSI(isp))
printf(" biu_conf1=%x", ISP_READ(isp, BIU_CONF1));
else
printf(" biu_csr=%x", ISP_READ(isp, BIU2100_CSR));
printf(" biu_icr=%x biu_isr=%x biu_sema=%x ", ISP_READ(isp, BIU_ICR),
ISP_READ(isp, BIU_ISR), ISP_READ(isp, BIU_SEMA));
printf("risc_hccr=%x\n", ISP_READ(isp, HCCR));
if (IS_SCSI(isp)) {
ISP_WRITE(isp, HCCR, HCCR_CMD_PAUSE);
printf(" cdma_conf=%x cdma_sts=%x cdma_fifostat=%x\n",
ISP_READ(isp, CDMA_CONF), ISP_READ(isp, CDMA_STATUS),
ISP_READ(isp, CDMA_FIFO_STS));
printf(" ddma_conf=%x ddma_sts=%x ddma_fifostat=%x\n",
ISP_READ(isp, DDMA_CONF), ISP_READ(isp, DDMA_STATUS),
ISP_READ(isp, DDMA_FIFO_STS));
printf(" sxp_int=%x sxp_gross=%x sxp(scsi_ctrl)=%x\n",
ISP_READ(isp, SXP_INTERRUPT),
ISP_READ(isp, SXP_GROSS_ERR),
ISP_READ(isp, SXP_PINS_CTRL));
ISP_WRITE(isp, HCCR, HCCR_CMD_RELEASE);
}
printf(" mbox regs: %x %x %x %x %x\n",
ISP_READ(isp, OUTMAILBOX0), ISP_READ(isp, OUTMAILBOX1),
ISP_READ(isp, OUTMAILBOX2), ISP_READ(isp, OUTMAILBOX3),
ISP_READ(isp, OUTMAILBOX4));
printf(" PCI Status Command/Status=%x\n",
Redo stuff for sparc64- primarily fix bus dma implementation. The endian stuff was right, but the busdma stuff was massively not right. Didn't really test on ia64 or i386- don't have the former h/w and my FreeBSD-current disk is unwell right now. Hope that this is okay. MFC after: 1 week
2002-04-02 23:36:14 +00:00
pci_read_config(pcs->pci_dev, PCIR_COMMAND, 1));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 18:12:29 +00:00
}
Reference in New Issue Copy Permalink